CogNet – An Experimental Protocol Stack for Cognitive ...mmlab.snu.ac.kr/courses/2007_advanced_internet/papers/CognetDesc.… · CogNet: Experimental Protocol Stack for Cognitive

CogNet: Experimental Protocol Stack for Cognitive Radio Page: 1

CogNet – An Experimental Protocol Stack for Cognitive Radio Networks and Its Integration with the Future Internet

1. Introduction: Recent “Moore’s law” advances in programmable integrated circuits have created an opportunity to develop a new class of intelligent or “cognitive” radios [1][2][3][4] which can adapt to a wide variety of radio interference conditions and multiple protocol standards for collaboration between otherwise incompatible systems. Such a cognitive radio would be capable of very dynamic physical layer adaptation via scanning of available spectrum, selection from a wide range of operating frequencies (possibly non-contiguous), rapid adjustment of modulation waveforms and adaptive power control. In addition, a suitably designed cognitive radio with a software-defined physical layer would be capable of collaborating with neighboring radios to ameliorate interference using higher-layer protocols. These higher layer coordination protocols could range from multi-node signal combining and coding methods to etiquette mechanisms all the way to fully collaborative multi-hop forwarding between radio nodes. Thus, suitably designed cognitive radios have the potential for creating a next-generation adaptive wireless network [5] in which a single universal radio device is capable of operating in a variety of spectrum allocation and interference conditions by selecting appropriate physical and network layer parameters often in collaboration with other radios operating in the same region. Such a “cognitive network” will lead to increased network capacity and user performance. Perhaps for the first time in the short history of networking, cognitive radios offer the potential for organic formation of infrastructure-less collaborative network clusters with dynamic adaptation at every layer of the protocol stack including physical, link and network layers [6][7]. While the development of cognitive radio hardware and software, especially at the physical layer, has received considerable attention, the question of how one transforms a set of cognitive radios into a cognitive network is much less well understood. As such, adaptive networks of cognitive radios represent an important but demanding research challenge for both the wireless and networking communities. The extreme flexibility of cognitive radios has significant implications for the design of network algorithms and protocols at both local/access network and global internetworking levels. In particular, support for cross-layer algorithms which adapt to changes in physical link quality, radio interference, radio node density, network topology or traffic demand may be expected to require an advanced control and management framework with support for cross-layer information and inter-node collaboration. At the wireless local-area network level, an important technical challenge is that of distributing and managing this inter-node and cross-layer information then using this control information to design stable adaptive networking algorithms that are not overly complex. At the global internetworking level, clusters of cognitive radios represent a new category of access network that needs to be interfaced efficiently with the wired network infrastructure both in terms of control and data. End-to-end architecture issues of importance include naming and addressing consistent with the needs of self-organizing network clusters, as well as the definition of sufficiently aggregated control and management interfaces between cognitive radio networks and the global Internet [8].

The proposed projects will develop the architectural foundation for the integration of cognitive networks into the global internet through two integrated research thrusts. The first is to identify broad architecture and protocol design approaches for cognitive networks at both the local network and at the global internetwork levels. This architectural study should lead to the design of control/management and data interfaces between cognitive radio nodes in a subnetwork, and between cognitive radio subnetworks and the global Internet. The second thrust is to apply these architectural results towards prototyping a comprehensive cognitive radio protocol solution (the CogNet stack) and use it for experimental evaluations on emerging cognitive radio platforms.

The CogNet protocol design will consider the requirements of cognitive radio networks which should be able to seamlessly adapt to changing spectrum usage conditions, physical topology/density, and end-user service requirements. A number of architectural issues will be examined as we try to identify an efficient and complete solution – these include definition of a global network control and management architecture with support for cognitive radios at the edge, support for various PHY modes including network coding and multi-node collaboration, protocols for dynamic spectrum coordination, discovery procedures for topology establishment and optimization, design of a flexible MAC layer protocol capable of dynamic changes in layer-2 network configuration, ad hoc collaborative group formation including support for address assignment and forwarding incentives for multi-hop networks.

The experimental component of this project aims to develop an open-source, Linux-based CogNet software protocol stack for use with emerging cognitive radio platforms (such as the GNU radio [9] used as the baseline for this project, the KU agile radio [10] or the Lucent/WINLAB network-centric prototype [11]), and make this software


available for community research. The prototype software will be validated in two steps: first as a local-area radio network with moderate numbers of cognitive radio nodes, and then as part of several end-to-end experiments using a wide-area network testbed such as PlanetLab [12] and Emulab [13] (and GENI [14] in the future). We believe that the proposed CogNet software prototyping, open-source distribution and experimental validation will be of significant value to both the wireless and the broader networking research communities. The wireless research community currently lacks a well-supported experimental cognitive radio platform and the CogNet software will enable further research on spectrum sharing algorithms, flexible MAC layer implementations, and cross-layer PHY-MAC optimization. The broader network community is trying to identify how future wireless networks will impact the architecture of the global internet and the CogNet protocol stack can be a key technology component for implementation of cognitive radio subnetworks planned for NSF’s GENI experimental system [8][15][16][17][18]. 2. Research Goals: Cognitive radio networks have a number of new and interesting capabilities:

• Spectrum agility and fast spectrum scanning over multiple frequency bands, providing local awareness of radio interference and the ability to change frequency bands on a per-packet basis

• Fast PHY adaptation, or the ability to change physical-layer waveforms on a per-packet basis and PHY collaboration modes such as network coding

• Spectrum etiquette protocol and dynamic spectrum policy implementation on a per-session basis • Fully programmable MAC layer, with the option of dynamic adaptation to meet service needs • Cross-layer protocol implementation capabilities based on integrated PHY, MAC, network algorithms • Ad hoc cluster formation, involving multi-hop packet forwarding among peer groups of radio nodes

Early work on cognitive radios has focused primarily on physical layer aspects, in particular software defined radios with some level of dynamic spectrum management capability [19][20][21][22][23][24][25]. There are also several ongoing projects aimed at developing cognitive radio hardware platforms [10][11] with capabilities such as frequency agility, software-defined radio and spectrum coordination. Some software work has also been done for cognitive radio platforms, most notably the GNU radio [9] which provides a development environment for building flexible SDR implementations. It is noted here that although these cognitive radio platform and PHY oriented projects represent a good start towards a technology base, very little has been done to date on top-down definition of a general protocol architecture for cognitive radio networks and its implementation in the form of an open-source protocol stack.

Figure 1: Overview of cognitive radio protocol architecture and its integration with the future Internet

The vision of adaptive wireless networks using cognitive radios is an appealing one, but one that depends critically on the existence of a protocol framework with control and management support for cross-layer collaboration between radio nodes (see Fig. 1). For example, collaborative PHY mechanisms such as network coding require control mechanisms to identify participating nodes, specify path diversity routes and eventually indicate (or download) applicable forward error correction algorithms. Similarly, for flexibility at the MAC layer, the control protocol should be able to distribute status necessary to infer current network topology and congestion conditions, together with the ability to coordinate changes in MAC functionality between a selected group of radio nodes. At


the network layer, radio nodes should be able to organize into voluntary ad hoc network clusters that agree to forward packets between themselves – this requires control protocol support for neighbor discovery, address assignment and routing table exchange. Cross-layer adaptation algorithms also require exchange of PHY and MAC level status information between nodes which participate in an ad hoc network cluster.

In view of the complexity and range of control and management functions required, it is becoming increasingly clear that we should partition the protocol functionality of the cognitive network in an explicitly-defined control plane and a data plane. The data plane protocol stack on each node contains the modules needed to support data communication between the wireless nodes and it exposes a set of controls for each module through an API. This API is used by a general and extensible “Global Control Plane (GCP)” to monitor, configure and adapt the data plane modules. The GCP is made up of several key components including a network management and control overlay, an interworking point where aggregated representations of the cognitive radio network state and control points are provided to the future Internet, and a radio bootstrapping function that can operate on a channel at the edge of the service band, and would have wider radio coverage than a typical service channel. In addition, nodes rebroadcast control packets using a controlled flooding mechanism thus providing global awareness to all cognitive radios within a subnetwork. The GCP protocol stack provides for distribution of control messages required to support various collaborative PHY, spectrum coordination, flexible MAC or ad hoc networking functions outlined earlier.

This project will identify baseline control protocols and APIs covering common requirements while also providing for future extensions as new adaptation concepts emerge. An important design issue is that of balancing the relatively diverse control requirements against complexity and network overhead. We observe that programmability at each layer of the protocol stack has the potential for building more robust wireless networks, but at the same time can lead to excessive complexity and possible instability if too many “control knobs” are turned at the same time. In addition, it will be important to identify cluster level and subnetwork level control aggregation mechanisms that allow for summarized interfaces between cognitive radio subnetworks, their peers and routers or hosts in the wired future Internet.

The focus of the proposed project is on the architecture of the cognitive network and its interface to the global future Internet, which is embodied in the control protocols and APIs. As a result, we plan to adopt a conservative approach in the data plane: we will only expose a modest number of control parameters at each protocol layer, and use simple distributed algorithms that do not depend on perfectly accurate control information. To focus our study, we define the following cognitive radio networking capabilities as the baseline and will design our control protocol implementation to meet their specific needs:

1. API for PHY layer adaptation (agility, change of modulation waveform), and support for collaborative PHY in the form of network coding.

2. Spectrum coordination protocols that facilitate dynamic sharing among radio nodes using mechanisms such as etiquette policies or spectrum server.

3. Autoconfiguration - bootstrapping and topology discovery - protocols that can be used to establish network connectivity after a cognitive radio device is turned on or enters a new service area.

4. Flexible MAC framework that permits programmable functionality capable of dynamic selection of channel sharing modes based on observed network conditions and traffic demands.

5. Network layer protocols that support service discovery, naming, addressing and routing in ad hoc wireless constellations, including features that provide economic incentives for collaboration. This network layer will interface with a cross layer network management overlay that can provide aggregated representations of the cognitive subnetwork state to the future Internet.

Taken together, the above set of cognitive radio capabilities will represent a major advance in the state-of-the-art, and should thus serve as a useful open-source software platform for further research within the community. The CogNet software stack to be developed in this project will thus enable a variety of cognitive networking experiments including both local area network experiments and more global end-to-end experiments when connected to experimental systems like PlanetLab and GENI in the future.

In the next section, we provide additional details on the protocol design and algorithmic aspects for each of the above major cognitive radio functions to be addressed in the project.


Figure 2: Network Coding

3. Cognitive Radio Functions Implementing the architectural vision in Fig. 1 will require development of several new capabilities based on an appropriate set of inter-module interfaces and protocols. The CogNet control plane must also be effectively integrated with the control and management of the future Internet backbone, taking into account factors such as hierarchical organization, aggregation of control information, and end-to-end service requirements. In this section, we outline design approaches and research issues for each of the key CogNet protocol capabilities listed at the end of Sec. 2.

3.1 PHY Adaptation and Network Coding: Communication networks today share the same fundamental principle of operation. Independent data streams may share network resources, but the information itself is separate. Routing, data storage, error control, and generally all network functions are based on this assumption. Network coding [26] breaks with this assumption. Instead of simply forwarding data, nodes may collaborate with each other to recombine several input packets into one or several output packets.

There are two main benefits of this approach: potential throughput improvements and increased robustness particularly for multicast and broadcast service scenarios. Robustness translates into loss resilience and facilitates the design of simple distributed algorithms that perform well, even if decisions are based only on partial information. In fact, successful reception of information does not depend on receiving specific packet content but rather on receiving a sufficient number of independent (recombined) packets. The principle of network coding is best explained with an example (from [26]) as shown in Fig. 2 below. Observe that network coding is a particular realization of collaborative PHY in which nodes cooperate to establish a topology for redundant path transmissions with intermediate coding stages carried out at certain forwarders. This particular example assumes that forwarding paths and coding to be done at each node can be pre-established for data streams A and B, and that intermediate nodes are capable of modulo-2 (or soft-decision linear) combining of packets in transit.

A number of theoretical results on network coding have been published over the past few years [26][27][28][29][32][33][34][35][36][37][38] of particular interest here being its robustness when introducing forwarding incentive mechanisms [37], but real-time experimental validation (see [30][31] for empirical evaluation in wired networks) has lagged behind because of the need for a programmable radio with cross-layer control capabilities. In particular, implementation of the coded multicast arrangement shown in the figure will require GCP support for path establishment, specification of forwarder functionality (routing and coding) and correlation of packets to be combined via a global packet ID (such as source address, destination address, port numbers and sequence number). In this project, we consider implementation of network coding as a representative example of collaborative PHY and plan to implement and validate one or more specific network coding scenarios using the GCP control mechanism along with programmable PHY functionality in CogNet.

3.2 Spectrum Coordination Protocols: A number of approaches have been proposed for improved spectrum sharing over the past decade. Notable methods being discussed in the technical and regulatory communities include property rights regimes [39][40][41][42], spectrum clearinghouse [43], unlicensed bands with simple spectrum etiquette [44][45][51], open access [46][47][48][49] and cognitive radio under consideration here. The distinctions between unlicensed spectrum regimes, open access and cognitive radio approaches are relatively subtle as they are all based on the concept of technology neutral bands to be used by a variety of services using radio transceivers that meet certain criteria. For example, cognitive radio may be viewed as a special case of open access or unlicensed regimes in which radio transceivers are required to meet a relatively high standard of interference avoidance via physical and/or network layer adaptation. The cognitive radio principles currently under consideration by the FCC and the research community span a fairly wide range of possible functionalities both at physical and network layers, as outlined in Fig. 3 below, which shows the protocol complexity and radio hardware complexity regimes for a number of possible coordination schemes.

The “agile wideband radio” scheme shown [50] at the lower right side of the figure is the most prevalent concept for cognitive radio in which transmitters scan the channel and autonomously choose their frequency band and modulation waveform to meet interference minimization criteria without any protocol-level coordination with neighboring radio nodes. We observe here that although autonomous adaptation of the radio PHY is the simplest method and requires no coordination standards, it suffers from serious limitations due to near-far problems (“hidden


nodes”) and the fact that interference is a receiver property while spectrum scanning alone only provides information about transmitters. Another simple technique is “reactive control” [53] of transmit rate/power, in which radio nodes do not have any explicit coordination with neighbors but seek equilibrium resource allocation using reactive algorithms to control rate and power, analogous to the way the TCP protocol reactively adjusts source bit-rate over the Internet. At a slightly higher level of protocol complexity, it is possible to use spectrum etiquette protocols [44][51] to improve coordination between radio nodes, using either Internet-based spectrum services or a common spectrum coordination channel at the edge of the shared frequency band.

Figure 3: Hardware and protocol complexity chart for potential cognitive radio approaches

The common spectrum coordination channel (CSCC) approach shown in Fig. 4 below used has been proposed [51][52] as a candidate mechanism for implementing spectrum etiquette policies. In this approach, each wireless device sends periodic beacons containing spectrum usage information so that neighboring nodes can avoid using the same frequencies, or if the network is too congested, can execute specified spectrum etiquette policies. It is observed here that the CSCC approach is similar in concept to the global control plane discussed earlier, and can thus be integrated into a GCP implementation as a subset of its overall functionality. In our earlier work [52] we have shown that CSCC implementations can achieve significant spectrum efficiency improvements with relatively simple etiquette policies when compared with radios with spectrum scanning and reactive agility.

Figure 4: Concept of CSCC for spectrum coordination

An alternative approach which can be used for spectrum coordination is the spectrum server. In this design, all radios use a spectrum management protocol to communicate with a centralized spectrum service within the future Internet. In our recent work in [54][55], we have examined the boundaries of system performance under the assumption that efficient access to spectrum can be resolved by an impartial “spectrum server” that can obtain information about the interference environment through measurements contributed by different terminals, and then

Hardware Complexity

Protocol Complexity (degree of

coordination)

Reactive Rate/Power

Control

Agile Wideband

Radios

Unlicensed Band

with DCA (e.g. 802.11x)

Internet Server-based

Spectrum Etiquette

Ad-hoc, Multi-hop

Collaboration

Radio-level Spectrum Etiquette Protocol

Static Assignment

Internet Spectrum Leasing

“Cognitive Radio”

schemes

UWB, Spread

Spectrum

Unlicensed band + simple coord protocols

“Open Access” + smart radios

Frequency

CH#N CH#N-1 CH#N-2 CH#2 CH#1 CSCC

: :

Ad-hoc net B Ad-hoc

net A

Ad-hoc Piconet

Master Node

CSCC RX range

for X

CSCC RX range

for Y

Y

X


offer suggestions for efficient coordination to interested service subscribers. There are many ways in which a spectrum server can coordinate a network of cognitive radios. Recent work in [54] has considered the role of the spectrum server in scheduling variable rate links while the work in [55] has considered the spectrum server’s role in demand responsive pricing and competitive spectrum allocation.

The presence of a spectrum server to regulate spectrum in a local area (or wireless subnet) raises several issues about its role in the end-to-end architecture of the system. These issues range from protocol interactions with the rest of the network architecture as well as spectrum and interference interactions with neighboring subnets. Topics to be considered in this study include design of the spectrum coordination protocol (either CSCC or spectrum server) to provide the right level of granularity in time and space, taking into account control overhead costs. For CSCC, there are design issues related to deciding what types of information to carry in the beacon, and whether this function should be integrated with radio bootstrapping and address assignment functions to be discussed later. When considering the spectrum server, an interesting issue is that of determining whether this constitutes a basic management & control plane service for the future Internet similar to DHCP given the expected growth in wireless computing devices, and if so, what type of distributed hierarchical implementations might be appropriate.

3.3 Radio Bootstrapping Protocols: Self-organization [56][57] is a key requirement for cognitive radio networks. The Global Control Plane provides a bootstrapping process that enables a radio node to be aware of itself, the surrounding nodes and current network status when it starts up. There are two phases in this process: (1) obtaining PHY parameters, reachability, and performance information by listening on a control channel or channel scanning; and (2) negotiation, during which the new node can negotiate with existing sub-networks for name/service discovery or performance optimization. The appropriate association and/or authentication process can then be initiated for the new node to join existing networks.

®

®

®

Internet

BootstrappingBeacons

CogNET-1CogNET-2

Name/ServiceDiscovery

MSG Type Payload Length Next Header

Source Identifier

Time Stamp

Center Frequency

1 8 16 24 32

PowerLevel

Modulation/Coding

BandwidthRateLevel Sub-Network Name

Available Services Figure 5: Bootstrapping and discovery process Figure 6: An example of the bootstrapping beacon format

The bootstrapping and discovery process is illustrated in Fig. 5, in which some cognitive radio clusters/sub-networks have access to the future Internet and some are “pure” wireless hierarchical networks. Bootstrapping beacons with information about current network status are broadcast within one hop on a specific control channel by existing active nodes in a periodic and opportunistic way. When a new node is starting up (or moving close to current sub-networks), it first listens for bootstrapping beacons on the control channel to obtain PHY and connectivity information to initialize its radio parameters by choosing proper operating frequency, transmit power, bandwidth/modulation/rate, etc. Service discovery information is obtained from the bootstrapping beacons, which allows the node to utilize higher levels of the Global Control Plane for network layer, naming, and cross layer management and control, and other services. Current nodes can provide performance information in the beacons, such as link speed, available network capacity, congestion indication, etc. In addition to common signaling channel, nodes can scan different traffic channels and listen to the beacons/send an association message to the best “cost” parent using the appropriate discovery metric.

Fig. 6 shows an example of the bootstrapping beacon format. The bootstrapping beacon is a low layer (PHY) messaging mechanism used to support neighbor discovery and determination of the logical topology. In addition, it is used to convey cross-layer information, for example, PHY parameters such as available frequencies, transmit power, modulation/coding type, radio bandwidth, bit-rate, and network name as well as available services. The beacon format should be kept as simple as possible aimed at letting the new node learn the necessary information about current network. In this project, we will consider various design options taking into account functional requirements such as topology discovery, self-organization and dynamic spectrum coordination.

3.4 Flexible MAC Layer: Wireless medium access control (MAC) protocols [59][61][62][63][65][67][68][69] tend to have different operating conditions in which they operate best. For example, RTS/CTS based MAC protocols


work well in hidden node scenarios, but also incur high overheads [58] when such collisions are rare. Cognitive networks provide us the opportunity to dynamically change MAC protocol [64] to suit both the needs of the applications running on the nodes as well as the properties of the environment around them. Our own previous work has shown that this form of adaptation is useful for higher layer protocols and recent preliminary work [59] has shown some promise in adapting MAC protocols as well [65]. Changing MAC behavior on-the-fly involves several research challenges as outlined below.

The first step in choosing the best MAC operating mode is to understand the propagation environment and traffic demand matrix of the involved nodes. The propagation environment must be measured/inferred using relatively sparse information and both the propagation environment and traffic matrix can change quickly over time. An added complication is that even when factors such as the propagation properties of the environment and the traffic demand matrix of the nodes are known exactly, choosing the best MAC protocol may be difficult. For example, a particular choice of MAC will rarely be best for all nodes involved. This requires carefully balancing the global utility of the system with fairness. In addition, the optimal choice may change dramatically over time. However, there may be considerable overhead in switching between different MAC behaviors and the system will need to manage such dynamic environments by balancing performance gains against mode switching overhead.

The different MAC layers that are supported may be inherently incompatible in their behavior. As a result, the system must ensure that all nodes are synchronized in any switch between MAC protocols. This ensures that the impact on reachability between nodes is minimal and short-lived. It also ensures that there is no oscillation between operating modes as different groups of nodes reach different decisions due to inconsistent information. Choosing a compatible MAC protocol ensures that a pair of nodes can communicate directly with each other. However, nodes using incompatible MAC protocols may be able to co-exist in an area much as existing incompatible wireless devices co-exist today. A key requirement of the flexible MAC is that it coordinates closely with any network topology management system. Nodes that are part of the same constellation must use compatible MAC protocols, while independent constellations may make independent optimization decisions.

The final issue we consider here is the design of the interface between the flexible MAC layer and the control plane. The control plane is responsible for reconciling different application requirements and network policies. It is also responsible for handling much of the coordination between nodes in the system. Our current belief is that the control plane is likely to operate on coarser time-scales. The flexible MAC protocol will adapt to more rapid changes in conditions. This partitioning of responsibilities results in an interface where the control plane may handle issues such as managing constellations and determining fairness policies but the flexible MAC chooses the current operating mode. We plan to explore alternate designs in this space by trading off GCP (global control plane) functionality with those implemented as part of the flexible MAC layer.

3.5 Network Layer Protocols: The network layer for cognitive radio networks will need to support a variety of applications including data and mobile real-time services (e.g., voice). To enable these given the variability of wireless communication, we propose the research and evaluation of a cognitive, multi-overlay network layer that can adapt based on sensing and learning of the cross layer environment, the communication model for the applications (e.g., one-to-one, one-to-many), and policy and security constraints. These latter issues are particularly important in a cognitive radio environment where opportunistic communications with previously unknown nodes may become common.

Figure 7: Cognitive Wireless Network with Multiple Network-Layer Overlays


The cognitive network layer will utilize both non-traditional approaches such as overlay-based mechanisms for communication within a subnet, as well as support the concept of supernodes [74] which will serve as a gateway between local network layers within the cognitive subnet as well as to the future Internet and its IP-based and overlay-based networks. Distributed hash tables provide a large number of optimization points, and may be tailored to the application at hand, inspiring the idea of multiple network layers tailored to specific applications or communication flow types. These might include routing overlays such as Virtual Ring Routing [70], which has shown promise in ad hoc network layer routing scenarios, application-tailored overlay structures such as topologically aware DHTs for group messaging applications[71], and DHTs that support rich queries [72][73].

Fig. 7 illustrates a cognitive radio network with mobile nodes and an interconnection to the wired core network through a supernode. Use of a particular overlay will be decided upon by cognitive techniques (e.g., case-based, expert system) based on factors including the wireless environment, application, and policy. Overlay 2, for example, might be a multicast optimized DHT-based network layer being used to support a messaging application. Overlay 1 might be IP or some other protocol useful in interacting with nodes on the future Internet.

3.5.1 Naming and Service Discovery: The cognitive radio network will support naming and addressing mechanisms that provide for self-organization, translation for global reachability, and merging / disconnection of cognitive networks themselves as well as with the wired network infrastructure. The approach is an extension of the concept of a supernode [74] to serve not only as a router between different geographically diverse cognitive radio networks, but to serve as a router between multiple network layers (e.g., DHTs, IP) within a cognitive radio subnet, and as a gateway to the future Internet. The identity of supernodes will be broadcast to new wireless nodes via the GCP‘s radio bootstrapping protocol. As a node joins a cognitive wireless network, layer 2 connectivity is established with neighbors and a join message for a service discovery [75] overlay can be sent to a neighbor node. This will allow the node to discover how to access all of the various services such as the network management overlay (discussed in more detail in 3.5.4.), service-specific network layers, or IP configurations. As nodes join the service discovery layer, the super node will expose the new nodes identifier to the wired network via the appropriate name service (e.g., DNS). If there are multiple supernodes with connectivity to overlapping wired networks, name based overlays such as i3 [76] can be used to specify the desired supernode gateway.

As an example of naming and addressing in this architecture, assume a new node in the cognitive wireless network is self configuring. In this scenario, it wishes to communicate with a node on the future Internet running IP. This node has discovered that the supernode is offering an IP service via the service discovery overlay, and receives an IP address and default gateway (the inside interface of the supernode). Note that an alternative approach might be to speak a non-IP network protocol over the subnetwork with translation performed at the supernode. In this case, however, the supernode exposes the wireless IP address to the outside world via DNS as nodeA.Rutgers.CogNet. A user wants to make use of the node A to receive inbound voice calls, and updates a mobility overlay such as ROAM [77] to let the outside world know that the user can be reached at nodeA.Rutgers.CogNet. An incoming call is initiated after the lookups are processed, and the packets are forwarded by the supernode over the cognitive wireless network. If a supernode loses connectivity with the wired network, it will continue to provide routing between the network layers of the disconnected cognitive wireless subnetwork. 3.5.2 Cross-Layer Aware Routing: Routing between cognitive radio nodes is influenced by control information from a number of sources. Nodes may obtain information about the application traffic, specified policies, link capabilities, MAC layer congestion status, and network reachability, and then decide on the most appropriate network layer to deliver the message to the peer. In order to support this routing decision, physical communication status and quality information from the cognitive wireless network will be exposed to participating nodes via Global Control Plane. It is observed here that MAC adaptation and routing decisions in presence of cross-layer information need not be independent, and (at the expense of complexity) more general integrated routing and MAC algorithms may be investigated for the cognitive network scenarios under consideration.

A key architectural issue is that of specifying control interfaces between routing protocols in the cognitive radio subnets and the wide-area internetwork. For scalability, these interfaces need to support suitable forms of aggregation and hierarchy in order to limit the amount of cognitive radio subnet specific information that needs to be visible to other routers in the future Internet. While routing of a packet from a corresponding host in the wired network would benefit from some visibility of network topology, MAC congestion and link quality within the cognitive radio network, this needs to be balanced against complexity and control aggregation considerations. In general, the simple abstraction of a “wireless link” does not apply to a multi-hop cognitive radio path when viewed from a host or router within the wired core. One potential solution for this is the concept of “dynamic topology


control” which has been proposed [8] as a possible abstraction that permits specification of topology and link quality in an integrated fashion to reflect the reality of the radio network. As part of this project, we will investigate alternative aggregated representations of cognitive radio network state and evaluate trade-offs between control granularity and end-to-end performance.

3.5.3 Forwarding Incentives in Cognitive Networks: A key design issue for ad hoc wireless networks including those formed by cognitive radios is that of creating voluntary collaborative groups with agreements to forward each others’ packets. This type of collaboration between radios makes the most sense as node densities increase to a point at which PHY-layer spectrum coordination cannot sufficiently resolve congestion problems in the system. Collaboration at the network layer via formation of ad hoc networks is considered to be a powerful mechanism for better utilization of spectrum, and can be associated with low-power radio transmission to nearby neighbors as well as higher bit-rate, better quality PHY links compared with direct connection to a distant receiver. In [84], it has been shown that mobility combined with forwarding can change the capacity per node of an ad hoc network from being not scalable [85] to being scalable. In [86][87], forwarding allows the exploitation of multi-channel diversity. [88] considers content distribution via single-hop multicast. In order to expedite data dissemination, a node also relays packets for other nodes if it has not done so for some time. A cooperative game approach is considered in [89][90], in which users are willing to stay in the network despite being required to forward for others.

While these works implicitly assume a willingness to relay data, in the context of heterogeneous users sharing spectrum, such a willingness to relay data needs to be incentivized via suitable protocol mechanisms. Using a microeconomic framework based on game theory [91][92], we have designed and analyzed a pricing algorithm that encourages forwarding among autonomous nodes by reimbursing forwarding costs accrued in terms of energy and lost opportunities for transmission of one’s own data while forwarding for others. Our results have shown that pricing with reimbursement appears to improve the network aggregate utility (or aggregate bits per Joule) as well as utilities and revenue compared to the corresponding pricing algorithm without reimbursement. Our work has also revealed that the nodes’ willingness to forward is greater when there is greater clustering of nodes in the network. Specifically, it has been observed that for large ratios of the average internodal distance to the smallest distance between the access point and any source node, the tendency to forward decreases.

While the above results have been derived in the context of a simple local area network formed of an access point and a set of wireless nodes, it bears asking the question of how this affects or (or in turn is affected) by an end-to-end network architecture. In a local area network context, the incentive based resource allocation algorithms explicitly take into account wireless transmission parameters such as energy constraints, radio channel quality, and throughput performance. In the presence on an overarching end-to-end architecture, the challenge is in understanding, improving and designing the algorithms for the cognitive radio subnet after taking into account interactions with reimbursement mechanisms within the wired Internet.

3.5.4 Network Management Architecture: Much network management infrastructure is still designed and deployed based on a hierarchical manager/agent architecture [78]. While there have been efforts to handle ad hoc wireless networks [79][80][81] and develop techniques for passively [82] and actively [83] monitoring overlays, these approaches all are missing a comprehensive approach for handling the variety of constraints that occur in a mobile dynamic RF spectrum wireless environment, and the unique fault localization environment associated with mapping faults that occur in a dynamic radio underlay into useful notifications for the changing substrate of overlays providing higher level services and applications.

In order to facilitate routing and cognitive RF functionality, the Global Control Plane

Network Management Information

Exposed ViaManagement AndService Discovery

Overlays



Overlays

Network Wide Cross Layer Interaction

Cognitive Radio & MAC Layers

Local Instance

of theGlobal ControlPlane



Overlays

Applications

Cognitive Networking Layer

IPRich Query

GroupMessaging

Node 1 Node 2 Node 3

Figure 8: Cognitive Wireless Network Management


will include a novel network management overlay to provide cross layer information from the entire subnet (see Fig. 8). Each node can expose their own cross layer information into the overlay, and access other nodes’ cross layer information as well, for a unique view into the hop by hop and end to end environment. The subnet wide management plane will be filled with and provide access to resource policies, node capabilities, link quality estimates, node availability, node performance metrics, fault localization, topology, and spectrum measurements that can be utilized via the GCP by protocol components from any network layer. Supernodes will be able to map topology and other network management data into evolving services in the future Interent such as topology services and information planes [78]. Applications themselves will also be able to make use of this information for fine tuning parameters such as codecs, packets sizes, timeouts, etc based on the cross layer information they see. We will explore the tradeoff of the additional overhead associated with maintaining and accessing the network management overlay to the additional performance and reliability gains from using this service.

4. CogNet Protocol Stack Implementation The Cognet protocol stack design is centered around the global control and management plane (GCP), while keeping the data path functionality relatively simple. The main building blocks for the CogNet protocol stack are:

1. PHY API for dynamic adaptation of physical layer parameters such as center frequency, bandwidth, modulation, and coding

2. Radio node bootstrapping and topology discovery protocol which is used to select initial operating parameters, obtain a network address and initiate communication with intended destinations

3. Flexible MAC control software and drivers which supports session-level setup of programmable MAC procedures in response to observed topology, traffic demand and performance

4. Generalized cross-layer routing framework which supports a range of possible ad hoc routing methods and types of link metrics with awareness of PHY and MAC layers

The project will leverage hardware platforms and software from the GNU open-source software defined radio project [9] as a baseline, with a modest investment to upgrade the software environment for specific CogNet requirements. The hardware platform used will be the Universal Software Radio Peripheral (USRP), which is a low cost, high speed USB 2.0 peripheral, specifically designed to enable the construction of software radios [93].

The USRP provides the minimal hardware required to make software radios a reality. The fundamental challenge with software radios is interfacing the signal processing software with the RF spectrum. The USRP solves this problem by providing a set of RF daughterboards to perform analog RF up and down conversion, 1 million gates of FPGA, typically used to convert to and from complex baseband, 4 high speed A/Ds (64 MS/sec 12-bit), 4 high speed D/As (128 MS/sec 14-bit) and a USB 2.0 controller chip. This hardware (see Fig. 9), along with free software, allows the USRP to convert an 8 MHz swath of RF spectrum to and from complex baseband and transfer it across the USB at up to 8 MS/sec complex. Everything is configurable by the user. The FPGA and USB controller chip firmware are loaded over the USB. All source code for the complete system, including Verilog for the FPGA is available freely under the terms of the GNU General Public License. In addition, although the USRP and its control software can be used without GNU Radio, the USRP is highly integrated into the GNU Radio code base [9]. The USRP board is expected to migrate to the higher performance “USRP2” board currently under development in year 2 of the project, increasing the bit-rate capability significantly via improved FPGA support. The USRP2, which will be used for networking experiments with CogNet in the later stages of the project, is expected to support bit-rates in the range ~10-50 Mbps depending on PHY processing requirements.

Blossom Research, one of the original developers of the GNU radio, will work on upgrades and software enhancements necessary for implementation of the CogNet radio stack in this project. In particular, incremental

Figure 9: The USRP software radio board


work for this project will focus on the definition of API’s for the PHY and software MAC. Both API's will most likely have the same “abstract shape” with the following capabilities:

• Allow higher level code to query capabilities of lower level via some kind of introspection. Expected to have the flavor of “I have these attributes, and for each one, this is the range of acceptable values”.

• Support dynamic reconfiguration of the lower layer, such as set and get properties. • Support some precise notion of time of reception, time to transmit. • Support sending and receiving annotated packets/frames. Example annotations include power level,

channel coding, time to transmit, time of reception, etc.

In addition to these API’s, GNU radio software enhancements for this project include development of a baseline programmable OFDM PHY with reasonably good flexibility in selecting bandwidth, modulation levels, etc. In addition, a simple software MAC (based on a random access protocol such as ALOHA or CSMA) will be developed for purposes of testing the CogNet stack, and can later be replaced with more advanced adaptive or multi-modal MAC protocols discussed in Sec 3.4.

5. Experimental Research Plan The CogNet software stack will be used to conduct several experiments in both wireless local-area network and end-to-end wide area network scenarios. At the first step, simple validation experiments will be conducted using simple channel emulator setups at CMU [98][99][100], WINLAB and KU. Wireless local-area experiments will initially be conducted using ~10-15 enhanced GNU radio boards (with capabilities outlined in Sec. 3 above), each connected to an ORBIT radio node in the radio grid testbed [94][95][96][97] at Rutgers in order to leverage experiment programming, execution and measurement capabilities in a controlled environment. In parallel, we will use the CMU wireless emulator for testing and evaluation: since it uses programmable channel models, we create increasingly more challenging channel models as the protocol software matures. For wide-area experiments, the plan is to interface one or more GNU radio clusters attached to ORBIT with a global overlay networking testbed such as PlanetLab. End-to-end experiments can then be conducted by interfacing the GCP to a wired network control and management overlay as discussed in Sec 3.5.4, while the data plane is connected to standard IP services. At a later stage, more comprehensive end-to-end layer 3 experiments at both data and control planes can be conducted using the proposed GENI infrastructure.

Specific research experiments aimed at different cognitive radio research topics to be conducted include:

• Evaluation of network coding algorithms for different multicast service scenarios in local-area environments. Performance criteria include estimation of control protocol overhead and data path robustness under specified channel impairment conditions. This experiment will be conducted using GNU radios connected to the ORBIT radio grid testbed with injection of controllable noise levels, later migrating to an outdoor GENI deployment.

• Comparison of spectrum agility with CSCC-based or spectrum-server based etiquette policies. This experiment involves dense deployment ~10 cognitive radio nodes running the CogNet stack and using the CSCC to implement spectrum etiquettes such as FCFS or time scheduling. This experiment will also be conducted within the ORBIT testbed framework to take advantage of available measurement and signal injection capabilities, and will later be ported to an outdoor deployment with larger numbers of nodes in an outdoor GENI deployment.

• Validation of protocols for self-organization and topology formation is cognitive radio networks. This experiment involves implementation of beaconing and subsequent bootstrap, discovery and address assignment protocols in a network with ~10’s of nodes running the CogNet stack. This experiment will also start out using ORBIT and will later migrated to GENI cognitive radio subnets for large scale studies.

• Experiments on network layer routing and forwarding in the cognitive environment, including a more traditional MANET/IP ad hoc environment, and exploring structured P2P forwarding environments, initially as an overlay, but also natively over the cognitive L2/L1. Models for microeconomic incentives in ad hoc + wired environments will also be evaluated using this framework. This will include internetworking with future wired networks planned for GENI.

• Evaluation of the flexible MAC by running experiments with different signal propagation models and traffic loads. Later in the project, experiments will be extended to PHY-MAC cross-layer optimization,


i.e. control algorithms can simultaneously control the PHY and MAC layers. These experiments will validate the PHY and MAC control APIs and the configuration and adaptation algorithms.

• Experiments on end-to-end network management of the cognitive system, including providing cross layer visibility throughout the network, and interfacing with the control plane to collect data and capabilities, and interfacing with future information plane and topology services delivered by GENI, etc. This will be focused on management of the network as opposed to management of possible experiments on the network. These experiments will be driven by services and applications that involve nodes both in the cognitive network and in the wired network.

6. Work Plan The major tasks in this project consist of the following:

1. Architectural research on cognitive radio networks, including control and data planes and their integration with the future Internet. This task includes consideration of cognitive radio networking requirements, definition of a protocol framework necessary to implement the global control plane, and interfacing of the GCP with the global network in terms of information exchanged, aggregation and hierarchies.

2. Protocol design and algorithmic research on PHY adaptation and collaboration. This task involves use of the GCP framework from task 1 to implement the control and data plane functionality required for network coding. A specific PHY adaptation protocol design will be designed and validated, and the related network coding algorithms specified and analyzed.

3. Protocol design and algorithmic research on self-organization, bootstrap and discovery. This task involves specification of a beacon for bootstrapping along with algorithms for topology discovery and network cluster formation. Initial validation will be carried out using simulation or emulation models, leading to preliminary software prototypes for inclusion in the CogNet stack.

4. Protocol design and algorithmic research on network-layer addressing, routing, cross-layer support and forwarding incentives. This task involves exploration of various alternative architectures for network-layer functionality in cognitive networks, with particular emphasis on defining control plane interfaces between the wired and wireless network. Potential schemes will be analyzed and a single general solution will be identified for implementation in the CogNet stack.

5. Cognitive radio platform preparation including driver and API support. This task involves building on the GNU software radio base to provide API’s for PHY and MAC layers for use with the USRP/USRP2 board. Software modules for OFDM PHY and an example software MAC will also be developed.

6. CogNet stack prototyping. This task involves design and software prototyping of the complete CogNet stack (both control and data) using the USRP/GNU platform developed in task 5. This protocol stack implementation will include the beaconing and global control plane (GCP) protocols along with API’s for control of PHY, MAC and network layers in the radio. Baseline spectrum sharing, adaptive PHY, adaptive MAC and ad hoc collaborative networking capabilities will be implemented using the GCP capabilities.

7. Experimental validation with moderate numbers of cognitive radio nodes as local-area network. Specific experiments include validation of protocols for network coding, spectrum etiquette, bootstrapping and discovery, and adaptive MAC using moderate numbers of GNU radios connected via the ORBIT grid. These experiments will later migrate to the cognitive radio subnet planned for GENI.

8. End-to-end experimental verification of cognitive radio subnetworks connected to the wide-area Internet using appropriate testbeds (initially PlanetLab, later GENI). This task involves end-to-end experimentation with the complete CogNet protocol stack developed in task 6 and validated in local environments in task 7. The focus of these efforts will be on validating control interfaces, aggregation and hierarchies necessary to support ad hoc cognitive radio clusters with cross-layer adaptation.

The project has been planned as a three year effort. The major project tasks and associated schedule, milestones and deliverables are given in Table 1 and Fig. 10 below.


Task# Work Item Resp.

Groups Project Deliverables

1 Architecture & requirements study

All Report on architecture & protocol design issues; research papers and evaluation results

2 Collaborative PHY and network coding

WINLAB Report and papers on network coding implementation using cognitive radios; specification and evaluation of protocols required and related algorithms.

3 Flexible MAC layer protocols

CMU Report and papers on flexible MAC layer implementations, along with evaluation of algorithms for MAC layer adaptation in various wireless scenarios. Preliminary simulations and software prototypes for soft MAC to be included in CogNet stack.

4 Network protocols and research

KU, CMU, WINLAB

Report & papers on network protocol design covering control & management interfaces, integration with wired networks, and ad hoc cross-layer operation. Preliminary simulations and software prototypes of routing used in the CogNet stack.

5 CR platform preparation & drivers

Blossom Research

USRP/GNU platform extensions necessary for OFDM and soft MAC based cognitive radio implementation. Includes radio API’s and PHY/MAC software.

6 CogNet Protocol Stack software

WINLAB CogNet protocol stack on USRP/GNU platform, including global control plane and protocols/algorithms for baseline adaptive PHY, MAC and routing. Open-source release of CogNet for use by research community and as a building block for GENI

7 Experimental validation with local-area network

All Experimental results validating protocol operation and performance of cognitive radio networks in local area scenario, using wireless network testbeds such as ORBIT.

8 End-to-end experimental validation

All Experimental results validating protocol operation and performance of cognitive radio networks in integrated wired-wireless scenarios, using wide-area testbeds such as PlanetLab and GENI.

Table 1: Project Tasks, Responsibilities, and Deliverables

Figure 10: Schedule & Milestone Graphic for Proposed Cognitive Radio Project

7. Education and Knowledge Transfer The project is expected to provide valuable training on software and systems aspects of wireless networks involving emerging programmable radio technologies. WINLAB/Rutgers, CMU and KU all have a history of ongoing student research projects on hardware design and network prototyping related to cognitive radio, software defined radios and ad hoc networks. This project is expected to strengthen this research area and lead to improved MS and PhD thesis opportunities for students with an interest in protocol design and implementation. WINLAB recently introduced a wireless systems course involving design and prototyping of hardware or software components, and the availability of a cognitive radio platform should provide a new dimension for experimentation on adaptive radio

USRP/GNU Platform development

CogNet Software Stack Development

USRP2 upgrade

Architecture & Requirements Study

Initial design V1 Board release

Year 1 Year 2 Year 3

V2 Board release

Paper design & evaluation

Report/spec

V2 software

Initial design V1 release V2 release GNU open-source

Collaborative PHY & Network Coding

Simulation results

Adaptive MAC Protocols

Network Layer Protocols

Paper design & evaluation Prototyping results & software

Paper design & evaluation Initial prototyping results

Net experimental evaluations

MAC experimental evaluations

PHY experimental evaluations

End-to-End Internet Integration Experiments (GENI)

End-to-end validation/demo

Adaptive wireless net prototype

Adaptive MAC demo

Network coding evaluation/demo


networks. The topic of cognitive radio will also be introduced into a graduate course on wireless systems, and the proposed project is expected to provide supplementary teaching materials and experimental options to strengthen the curriculum. In addition, KU has developed a short course on software radio architecture and implementation.

WINLAB, CMU and KU ITTC each have active industry outreach program involving technical seminars, collaborative research, visits by technical staff from sponsor companies, etc. The proposed project is expected to create several opportunities for joint work with industry sponsors developing next-generation radio technologies. WINLAB industrial sponsors have recently shown a great deal of interest on the topic of spectrum management and cognitive radio, and this project is expected to provide useful design and experimental insights of interest to these companies.

8. Impact Statement This proposal has the potential for significant scientific and societal impact. Cognitive radio technology represents an important area for public policy at the FCC, and is currently the subject of a notice-of-inquiry that may lead to assignment of new frequency bands for this purpose. This project is aimed at developing state-of-the-art cognitive radio networking platforms of potential value to the research community for the purpose of evaluating proposed spectrum coexistence methods. If successful, the output of this effort will be a cognitive radio protocol software stack that, along with a suitable hardware platform, can be used by research groups investigating the merits of the cognitive radio approach and attempting to understand protocols and algorithms that promote coexistence. The proposed CogNet protocol stack will also provide a key building block for experimental research on the future Internet using platforms such as the GENI system proposed by NSF CISE.

End-to-end evaluation of cognitive radio subnets integrated with the wired network should provide important insights for the design of protocols for the future Internet. In particular, this project should help to clarify some open issues related to integration of naming and addressing for autonomous wireless clusters, and should identify practical methods for aggregating and conveying cross-layer information across the global network without adding excessive overhead or complexity.

9. Results from Prior NSF Support Dipankar Raychaudhuri has 25 years research experience in networking and communications, and has led several large industrial research projects including prototyping and field trials of an early 5 GHz WLAN (WATMnet) system. After moving to Rutgers in 2001, Prof. Raychaudhuri began NSF-funded work [101] on etiquette protocols and coordination algorithms for spectrum sharing in unlicensed bands. His current research is also directed towards protocol design and prototyping of next-generation wireless systems and sensor networks [102][56][57][103], and he is currently serving as PI on a NSF NRT collaborative project entitled “ORBIT: Open Access Research Testbed for Next Generation Wireless Networks” (NSF award# ANI-0335244, 2003-07) [94]. He is also co-PI of a collaborative Lucent/WINLAB/GA Tech NSF project (award# CNS-0435370, 2004-08) [11] aimed at development of a network-centric cognitive radio hardware platform.

Narayan Mandayam has been supported by the NSF grants [104][105][106][107][108][109], an equipment grant [110] and by the NSF CAREER award [111]. The grant [105] enabled studies on issues related to: Implications of fading channels on physical layer performance, performance limits of error-correction coding for wireless data over correlated fading channels; QoS guarantees for wireless data via flexible software defined radio (SDR) architectures [19][112][113], integrated access control with detection [114][115], and hierarchical control methods for resource allocation [116]. The grant [106], supported studies on power control for wireless data services. Two other NSF grants that both concluded in summer 2003 [104][111], resulted in channel measurements and modeling for infostations [117][118][119][120] and understanding of pricing transmitter power in the context of mediation and information theory [121]. The ITR grant [107] supported work on microeconomic principles related to spectrum policy and resource allocation. The grant [108] is studying cognitive radio technologies for open access to spectrum [54][55] and the grant [109] is investigating uncoded communications for fast feedback in multiple antenna systems.

Predrag Spasojevic, as part of a three university collaboration involving eight investigators, is involved in spectrum regulation under NSF grant [101] whose goal is to develop a general framework for understanding cooperation in unlicensed band wireless networks by studying economics/regulatory-based and protocol/signal-processing based communication techniques. He has also been involved in studying cognitive radio technologies for open access to spectrum under the NSF grant [92]. Prof. Spasojevic's work has been directed towards coding for collaborative wireless and wired communications networks [127][128][129][130], various signaling techniques in unlicensed


bands including ultra-wideband modulation [122][123], and limits of approaches to adaptive transmission and coding [124][125][126][131][132][133][134][135][136][137].

Ivan Seskar has worked on experimental infrastructure and prototyping on a number of NSF grants at Rutgers WINLAB, including development of an FPGA/DSP-based SDR radio in the mid-1990’s [19], prototyping of a high-speed OFDM radio for use with an Infostation, and implementation of the radio nodes and radio grid emulator in the ORBIT testbed currently under construction [94]. He is a co-PI of the Lucent/WINLAB/GA Tech NSF project (award # CNS-0435370, 2004-08) [11] aimed at development of a network-centric cognitive radio hardware platform.

Peter Steenkiste is a co-PI on the ITR “Proactive Self-Tuning Systems for Ubiquitous Computing” (CCR-0205266, September 2002 start). One result from this project is the development of an integrated Contextual Information Service that gives pervasive computing applications access to contextual information through an SQL-like interface [138]. We have developed a number of context-aware applications that use the CIS [139][140] and the CIS has formed the starting point for several student projects [141][142][143]. Peter Steenkiste is also a PI on the NSF project “Using Signal Propagation Emulation to Understand and Improve Wireless Networks” (CNS-0434824, September 2004 start). The project is developing a wireless network emulator that uses FPGAs to implement fully programmable real-time channel models that are used by real wireless devices, allowing fully repeatable and controllable wireless experiments [99][100][144]. We currently have an operational 4-node system that supports the full 2.4 GHz ISM band we will be increasing the number of nodes in the coming months.

Srinivasan Seshan received an NSF CAREER grant (ANI-0092678) starting in June 2001. The grant, titled “Towards an Efficient Ubiquitous Computing Infrastructure”, funds research into multi-modal protocols, network protocols that take on very different behaviors to adapt to gross changes in operating conditions. The work on this proposal has led to new routing and transport protocol designs [145] and new sensor network protocols [146]. This grant also supports work on upgrading nodes in a ubiquitous system, which appeared in HotOS 2003 [147]. Seshan is also co-PI on the NSF project “A Measurement-Driven Approach to Internet Protocol and Systems Design” (CNS-0435382, September 2004 start). The goal of this project is to use large-scale measurement studies of the Internet to help drive the development of Internet protocols. This project has already had a number of significant results on network routing [148], DNS deployment [149][150] and reliable system design [151][152]. Finally, Seshan is also co-PI on “Synthetic Reality: Physically Rendering Dynamic 3D Objects from Programmable Matter” (CNS-0428738, September 2004 start). Seshan’s work on this project explores the challenges of networking large numbers of programmable robots. Finally, Seshan and Steenkiste are co-PIs of “Self-managing Wireless Networks: Bringing Order to Chaotic Wireless Deployments” (CNS-0520192, September 2005 start). This project explores the issues of auto-configuring wireless devices in unplanned, unmanaged wireless environments [153].

Joseph B. Evans has over 20 years research experience in networking and communications, has led several large research projects including prototyping and testbeds for high-speed and wireless networks, and served as an NSF program officer from 2003-2005. He was a co-investigator on “Establishment of a Lightwave Laboratory for Applications Focused Research”, NSF EPS-9632617, which equipped and provided the staffing for a state-of-the-art lightwave communications research laboratory at the University of Kansas, a co-PI on “Ambient Computing Environment”, NSF EIA-9972843, which provided the basis for several major efforts in pervasive computing systems, “The Future of Spectrum: Technologies and Policies Workshop”, NSF ANI-0332249, an interdisciplinary workshop which discussed radio spectrum issues with researchers and policy makers, and was PI on “Flexible Wireless Systems for Rapid Network Evolution”, NSF ANI-0230786, which developed agile wireless networking technologies.

Eric Blossom Eric Blossom is the founder of the GNU Radio project [9] and overall architect of the system. He has 25 years experience in the design, development and deployment of large scale software & hardware systems. For the past 10 years his work has focused on wired and wireless communications. His current research exploits the capabilities of software radio as a foundation for building cognitive radios. His work on GNU Radio and the “Universal Software Radio Peripheral” (USRP) [93] is supported by the NSF NeTS-ProWIN project entitled “An Open, Low Cost, High Quality Software Radio Platform and Testbed” (NSF award #0435485, 2004-2008).

Documents

CogNet – An Experimental Protocol Stack for Cognitive ...mmlab.snu.ac.kr/courses/2007_advanced_internet/papers/CognetDesc.… · CogNet: Experimental Protocol Stack for Cognitive