1

Design of Fieldable Cross-Layer Optimized

Network using Embedded Software Defined Radios:
Survey and Novel Architecture with Field Trials
Jithin Jagannath, Anu Jagannath, Justin Henney, Tyler Gwin, Zackary Kane, Noor Biswas, Andrew Drozd
Marconi-Rosenblatt AI/ML Innovation Lab,
ANDRO Computational Solutions, LLC, Rome NY
{jjagannath, ajagannath, jhenney, tgwin, zkane, nbiswas, adrozd}@androcs.com
arXiv:2201.09454v1 [cs.NI] 24 Jan 2022

Abstract—The proliferation of wireless devices and their ever drive towards the software virtualization of these devices and
increasing influence on our day-to-day life is very evident and networks to provide much-required flexibility and adaptability
seems irreplaceable. This exponential growth in demand, both to the growing needs of each application. The inability to adapt
in terms of the number of devices and Quality of Service (QoS)
had spawned the concept of cross-layer optimization several years the network operations depending on varying requirements and
ago. The primary goal of the cross-layer approach was to liberate dynamic deployed scenarios due to black-box style wireless
the strict boundary between the layers of the traditional Open stack leads to inefficient resource usage and suboptimal per-
Systems Interconnection (OSI) protocol stack. The objective was formance. Dynamic network control and agile management
to enable information flow between layers which then can be (user scheduling, radio resources, mobility management) are
leveraged to optimize the network’s performance across the layers
mitigating the challenges due to scarce resources while meeting the envisioned benefits of the next-generation wireless net-
QoS demands. The initial decade focused on establishing the work paradigms such as software-defined networking (SDN),
theoretical feasibility of this revolutionary concept and gauging network function virtualization (NFV), and multi-access edge
the effectiveness and limits of this idea. During the next phase, computing (MEC) [8]. Therefore, a reconfigurable radio stack
the advent of software defined radios (SDR) accelerated the that can be transformed based on diverse applications as well
growth of this domain due to its added flexibility. Even with
the immense interest and progress in this area of research, as to adapt to the dynamic wireless conditions is favored.
there has been a gaping abyss between solutions designed in Traditionally, the strictly layered architecture proposed by
theory and ones deployed in practice. To establish this, we the open systems interconnection (OSI) reference model has
first present an elaborate survey of the cross-layer protocol been the prevalent design for a majority if not all modern
stack literature with emphasis on their maturity scale. Next, networking architectures. This is strict in the sense that they
we briefly discuss how a commercial off-the-shelf (COTS), low
SWaP (Size, Weight, and Power) embedded SDR (e-SDR) was are designed to maintain only a limited interface between
transformed into a standalone, fieldable transceiver. Thereafter, the neighboring layers [9]. Realizing the deficiencies in this
we provide the software design ethos that focuses on efficiency layered architecture, the cross-layer optimized approach has
and flexibility such that the optimization objectives and cross- been proposed over the past decade to maximize the utilization
layer interactions can be reconfigured rapidly. To demonstrate of scarce resources by "erasing" the strict boundaries between
our claims, we provide results from extensive outdoor over-the-
air experiments in various settings with up to 10-node network various layers of the protocol stack. In other words, any
topologies. The results from the field trials demonstrate high attempt to violate the OSI reference model is considered a
reliability, throughput, and dynamic routing capability. To the cross-layer design [10]. While there are abundant solutions
best of our knowledge, this is the first time in literature, a proposed in literature [9], [11]–[14], the majority of it is
COTS e-SDR has been leveraged to successfully design a cross- limited to simulations that may have strong assumptions and/or
layer optimized transceiver that is capable of forming an ad hoc
network that provides high throughput and high reliability in a do not consider all the hardware constraints and rigidness that
ruggedized, weatherized, and fieldable form factor. may be encountered during a real-life deployment. During
the next phase of advancement, the advent of software de-
fined radios (SDR) provided the much-needed impetus to this
I. I NTRODUCTION concept providing the flexibility to implement novel cross-
In recent years, we have seen explosive growth in the layer architectures. This enabled some of these efforts to be
number of wireless devices that have become an inevitable extended to hardware-based testbed evaluations. In most cases,
component of our daily lives. This includes everything such these efforts still used one or more dedicated (non-embedded)
as the connected devices in our smart homes, cellular net- host computers to implement the solutions which were then
work, the entire concept of Internet-of-Thing (IoT) networks connected to SDRs.
controlling manufacturing, monitoring smart grids, space com-
munications, underwater networks, tactical networks, among
others. As we move from 5G (5th Generation) to 6G (6th A. Motivation, Problem Statement, and Contribution
Generation), the need to optimize the scarce resources is Motivation: Even with the advances discussed above, to the
becoming evident and inevitable [1]–[7]. We have also seen a best of our knowledge, there does not exist a ruggedized and
 2

fieldable SDR with a comprehensive cross-layer optimization software protocol stack also provisions end-user-defined
capable software module implemented on an embedded ARM applications.
processor. The main reason for this is the various hurdles • We have conducted an elaborate outdoor experimental
that are associated with developing the solution from theory evaluation that includes, peer-to-peer setting, line net-
to effective hardware deployable software. In this article, work, 5-node topology, and 10-node topology.
we aim to demonstrate that this challenge can be overcome • The experiments demonstrated long-range, high through-
using a Commercial off-the-shelf (COTS) embedded SDR (e- put, and reliability in all the settings. The real-time adapt-
SDR) and a novel efficient software architecture. Therefore, ability and dynamic routing capability of the cross-layer
in this article, we present the first, completely stand-alone, optimized network have been demonstrated. Finally, we
ruggedized, and fieldable cross-layer optimized solution built also demonstrated how the network can handle multiple
using a low SWaP (Size, Weight, and Power) e-SDR. sessions without degradation.
Problem Statement: Driven by our motivation, we define The rest of the paper is organized as follows, in Section
a specific problem statement that is considered in this work. II, we survey the recent works in the cross-layer domain
Design a fieldable e-SDR with an energy-aware cross-layer with an emphasis on the maturity of the proposed solutions.
optimized protocol that aims to maximize network lifetime Next, in Section III, we discuss the system design from the
for enabling telemetry collection of tactical test and evaluation hardware and software point of view. We elaborate on the
ranges. network topologies and various test objectives for the field
The primary objective of this work is the transition of trials that showcase the reliability, throughput, and dynamic
theoretical cross-layer concepts that have been shown to be routing capabilities of the fieldable e-SDR in Section IV.
effective in simulations to a fieldable hardware entity. Our Finally, we conclude the article in Section V.
implementation principle has been to facilitate rapid reconfig-
uration of network objectives with only a few lines of code II. S URVEY OF C ROSS - LAYER A PPROACHES
thereby enabling a truly software-defined radio that evolves
Cross-layer approaches have been explored as a novel
with the growing requirements. To accomplish this, in the
solution to address a wide range of problems in wireless
extension to our prior work [15], we first discuss the hardware-
communication due to its perceived benefits in sustaining
level modifications required to customize the baseline e-SDR
communications in a dynamic and constrained environment
into a fieldable solution. Next, we outline how the software
[9]–[14], [16], [17]. This includes tactical network [18]–[20],
protocol stack was designed and implemented on a compu-
commercial terrestrial network [21]–[24], space networks [25],
tationally constrained ARM processor. To demonstrate the
[26], acoustic underwater networks [27]–[32], and even in the
feasibility of the implementation and the designed transceiver
upcoming visible light communication networks [33]–[37].
in terms of throughput and reliability, we performed extensive
These solutions are designed for various objectives such as
outdoor experiments with up to 10 nodes in the network. The
optimizing throughput and/or latency [38]–[41], fairness [18],
proposed architecture and design principles can be leveraged
[42], [43], energy consumption [19], [22], [44]–[46], resource
to implement and mature several of the novel cross-layer
management [24], [34], efficient multipath TCP [21]. The
optimized solutions to meet the evolving needs of both tactical
goal of this survey is not to cover the entirety of cross-layer
and commercial communication systems. We hope and believe
approaches, but rather to demonstrate the gamut of the areas
the unique cross-layer design methodologies and extensive
of application for cross-layer approaches to emphasize the
outdoor evaluations (field trials) will serve as an impetus for
impact and importance of these solutions. In contrast to other
the maturation of novel cross-layer solutions.
surveys, in this section, we specifically focus on reviewing
Contribution: The key contributions of the article are as
the maturity of these state-of-the-art cross-layer solutions to
follows;
expose the absence of a deployable e-SDR based solution. We
• To place the contribution in perspective, we survey the
divide the discussion into two broad categories, works that are
domain of cross-layer optimization with an emphasis on
limited to simulations and others that have provided some form
the maturity of the solutions. specifically, we discuss how
of preliminary hardware testbed-based evaluation to validate
the majority of the works in this domain is restricted to
their work.
either simulations or preliminary in-lab hardware analy-
sis.
• We provide a brief account of hardware components and A. Initial Proof of Concept in Simulation
their configuration that were used to customize the COTS As in any novel research, simulations are the obvious first
e-SDR to ensure a standalone, portable solution that choice to establish feasibility and performance gains over
would provide extended range and is capable of running existing approaches. To this end, the majority of the works
cross-layer optimized geographical ad hoc routing. that propose cross-layer optimization have been limited to
• Embedded software-defined cross-layer protocol stack simulations [18]–[22], [38]–[40] due to the challenges, time,
implementation with a clearly laid out explanation show- and effort it takes to evaluate them on hardware testbed.
ing all steps from the design concept to software imple- Most works in the cross-layer domain rely on MATLAB
mentation. [20], [38], [45], [47], NS2/3 [22], [42], [43], [46], OMNET
• Additionally, we provide details of the implementation of [19] or similar wireless network simulators. In most cases,
graphical user interface (GUI) as an example of how the these simulations are executed under various assumptions
 3

and/or are abstracted from the physical layer (PHY) of the Short Range Communication (DSRC) based vehicular com-
protocol stack and are restricted to packet-level simulators. munication networks. To this end, they introduce cross-layer
These assumptions and abstraction imply several intricacies optimization modeling for real-time, on-road multimedia ser-
of real-world deployment (operating channel conditions, com- vices and present a throughput-maximal framework that is
putational resources, RF frontend capabilities, flexibility, and designed for varying fading channel in mobile nodes with high
associated latency/overhead) are overlooked or set aside to be velocities. The authors claim the performance in the simulation
handled in the future which often thwarts the maturation level provided evidence for performance improvement which could
of cross-layer solutions. be relevant even beyond vehicular networks.
Since there are a large number of solutions proposed in In [49], the authors propose a cross-layer multi-path routing
this domain and evaluated in simulations, we categorize them approach that utilizes non-correlated and node-disjoint paths
based on the objective of the cross-layer approach. At the able to concurrently transport multimedia content from sources
same time, we would like to point out that these are not to the destination by accounting for contextual information.
strict classifications as many cross-layer approaches take into To accomplish this the nodes utilize cross-layer interaction
account multiple metrics either in their objective function or between the MAC and the network layer to tune their wake-up
constraints. Hence, a single work could arguably fall into schedules and determine the disjoint paths. The work is limited
multiple categories simultaneously. Therefore, in most cases, to simulation but the authors have observed performance
the categorization is based on at least one of its core objectives. improvement in terms of throughput, latency ad energy usage.
The main purpose of the categorization is to make it easier for 2) Minimizing Delay or Latency: A key factor that impacts
readers to assimilate all the content and show the widespread cellular networks and wireless mesh networks is the delay or
use of these techniques. latency experienced by the network. Latency is often a key
1) Maximizing Throughput: From the very early days, Quality of Service (QoS) metric required by end customers
throughput has been a central metric for any kind of commu- relying on wireless services for the operations. Accordingly,
nication device. The evolution of cellular networks, WiFi, or cross-layer approaches are proposed to minimize end-to-end
any other radio access technology is often driven by the desire delay for a multihop wireless network in [50]. The authors
to drastically improve the throughput of the network. Hence, propose two cross-layer schemes that they refer to as, loosely
it is not surprising to see a number of cross-layer optimized coupled cross-layer scheme and a tightly coupled cross-layer
solutions have been developed for this exact purpose. scheme. In the loosely coupled cross-layer scheme, routing is
Cognitive radio networks (CRNs) are referred to as net- computed first and then the information of routing is used
works that are capable of sensing and dynamically accessing for link layer scheduling whereas, in the tightly coupled
the spectrum of interest when it is idle and not being used by scheme, routing and link scheduling are jointly solved in
primary users. Primary users are entitled to a license to use one optimization model. They showcased the superiority of
the spectrum at the highest priority. The CRN can be seen their cross-layer approaches over traditional non-cross-layer
as secondary users who are allowed to access idle parts of approaches through simulations.
the spectrum without disrupting the primary user’s activities. One of the key objectives of 5G networks is to support ultra-
These are general definitions and could be more stringently reliable and low-latency communications (URLLC). Cross-
defined from case to case in actual deployments. A distributed layer approaches have been leveraged recently to enable
cross-layer approach to maximize the network throughput of URLLC in radio access networks by devising a resource
CRN was proposed in [38]. The goal was to perform oppor- allocation policy and a proactive packet dropping strategy [40].
tunistic spectrum access and dynamic routing algorithm and In this work, the authors optimize the packet dropping policy,
was referred to as ROSA (ROuting and Spectrum Allocation power allocation policy, and bandwidth allocation policy to
algorithm). The algorithm enabled individual nodes to perform minimize the transmit power under the Quality of Service
joint routing, dynamic spectrum allocation, scheduling, and (QoS) constraint. The feasibility of the proposed approach is
transmit power control. The proposed solution is shown to validated using simulations.
outperform approaches that considered only spectrum allo- In [51], the authors identify that the end-to-end delay perfor-
cation or routing individually using a packet-level MATLAB mance has a complex dependence on the higher-order statistics
simulator. of cross-layer algorithms, and hence optimization-based design
In the case of a CRN employing Code Division Multiple methodologies that optimize the long term network utilization
Access (CDMA) at the physical layer, a cross-layer algorithm are not ideal for delay-aware design. To overcome this, the
to maximize throughput was proposed using a new spread- authors design a delay-aware joint flow control, routing, and
spectrum management paradigm using [39]. The algorithm scheduling algorithm for multihop wireless networks to max-
referred to as ROCH (Routing and cOde division CHannel- imize network utilization. The proposed cross-layer approach
ization) proposed jointly-designed routing and code division utilizes a regulated scheduling strategy based on a token-based
channelization similar to ROSA with the added dimension of service discipline, for shaping the per-hop delay distribution.
codeword optimization. This work also relied on packet-level Through simulation, the authors show optimal network utiliza-
simulators for the initial proof-of-concept. tion and better end-to-end delay performance for the proposed
In [48], the authors propose a cross-layer approach that joint flow control, routing, and scheduling algorithm.
utilizes parameters from Medium Access Control (MAC) A cross-layer design is proposed in [47] using virtual queue
and PHY layer of the IEEE 802.11p protocol in Dedicated structures to guarantee finite buffer size or worst-case delay
 4

performance. The algorithm solves a joint congestion control, authors point out that the shortcoming in the existing CMT
routing, and scheduling problem in a multihop wireless net- solutions deployed at the transport layer is due to uncertainties
work while ensuring average end-to-end delay constraints per- at the lower layer (for example caused by variations of
flow and minimum data rate requirements. Through MATLAB the wireless channel). They also point out that CMT-based
simulations they illustrate a tradeoff between the throughput video transmission may unfairly use excessive bandwidth in
and average end-to-end delay bound while satisfying the comparison with the popular TCP-based flows. To overcome
minimum data rate requirements for individual flows. these issues, the authors propose a fairness-driven cross-Layer
DRS (Distributed Deadline-Based Joint Routing and Spec- SCTP-based Concurrent Multipath Transfer approach (CMT-
trum Allocation) [20] was introduced to mitigate the last CL/FD). CMT-CL/FD monitors and analyzes path quality,
packet problem that has been shown to be a shortcoming of which includes wireless channel measurements at the data-
queue length-based backpressure scheduling algorithms such link layer and rate/bandwidth estimations at the transport
as ROSA. In this work, the authors proposed the use of a layer. They also propose a window-based mechanism for
virtual queue such that the virtual queue length was designed flow control to obtain an acceptable tradeoff between delivery
to mitigate the effects of the last packet problem without fairness and efficiency. They utilize simulations to demonstrate
substantial degradation of the throughput performance. The the effectiveness of the proposed approach over traditional
virtual queue length considered parameters such as length of solutions.
the packet, remaining lifetime, and estimated time to the des- 4) Minimizing Energy Utilization: In recent years, green
tination. The distributed implementation of DRS was shown communication protocols have gained substantial interest. The
to outperform ROSA in terms of effective throughput and term "Green" in most cases refers to communication proto-
reliability using a packet-level simulator. The reason we have cols that aim to reduce the energy consumption of wireless
this work categorized under latency is due to the definition of networks. One can imagine the implications and benefits of
effective throughput used in this work. Only packets received minimizing energy consumption in various scenarios in the
at the destination before the deadline contributed to effective context of operating cost, mitigate the need to re-deploy nodes
throughput which essentially is trying to ensure packet delay that rely on battery, and its implication on climate change. Due
constraints. The simulations in MATLAB demonstrate that to these reasons, cross-layer optimization has been leveraged
DRS outperforms ROSA both in terms of effective throughput as a tool for minimizing energy consumption or maximizing
and reliability. the network lifetime. Typically, network lifetime is defined as
3) Fairness: Two cross-layer algorithms, a dual-based al- the total operation duration of the network before the first node
gorithm, and a penalty-based algorithm are proposed in [18] depletes its energy resources.
to solve the rate control problem in a multihop random access Authors propose a joint cross-layer optimization scheme
network. Both algorithms can be implemented in a distributed that considers modulation, power control, and routing to
manner, and work at the link layer to adjust link attempt maximize the energy efficiency of a wireless network [44].
probabilities and at the transport layer to adjust session rates. The constraints considered in their optimization problem in-
Their convergence and effectiveness of the proposed solution clude Bit Error Rate (BER) and the requested data rate.
were established using simulations. Each node in their multihop wireless network is capable of
In [42], the authors investigate the issue of fairness be- transmitting seven different modulations schemes. They use an
tween IEEE 802.11-compliant wireless local area networks event-driven, link-level simulator implemented using C++ to
(WLANs) stations using Transport Control Protocol (TCP). demonstrate improvement over Ad hoc On-demand Distance
Specifically, focusing on send/receive TCP traffic in WiFi Vector (AODV) routing protocol.
hot spots, they show WiFi hot spot provides more service As discussed previously, CRN aims to use an idle (unused)
to the wireless sending station compared to the receiving portion of the licensed spectrum to access the channel in a
stations. This unfair service implies that the wireless sending dynamic manner. This technology is often critical for IoT
stations dominates the use of network bandwidth restricting devices that may be energy-constrained. To overcome this
the access to the receiving stations. To mitigate this unfair challenge, the authors of [45] propose a cross-layer approach
service issue, the authors propose a cross-layer feedback that jointly considered the modulation order at the physical
mechanism in which the MAC layer at the access points layer and the backoff probability from the MAC-layer to
measures the per-station channel utilization and system-wide minimizing the energy consumption in the CR-based green
channel utilization to calculate the channel access cost. Next, IoT networks. The constraints considered in this work are to
the TCP senders use the cost in order to assure per-station ensure IoT delay guarantees, licensed primary radio channel
fairness and to maximize channel utilization simultaneously availability, and PR user activities. The authors use MATLAB
hence invoking a cross-layer interaction between the data link simulations to evaluate the performance of the proposed so-
and transport layer. Their simulations in NS2 demonstrate lution. The simulations show that the proposed cross-layer
that the proposed approach outperforms existing schemes with design is able to reduce energy consumption while achieving
respect to fairness, delay, and channel utilization. the delay requirements in the network compared to traditional
Stream Control Transmission Protocol (SCTP)-based Con- approaches that operate independently at the physical layer or
current Multipath Transfer (CMT) was designed to improve data link (MAC) layer.
the wireless video delivery performance with its parallel One of the key challenges of Wireless Sensor Network
transmission and bandwidth aggregation features. In [43], the (WSN) is ensuring energy efficiency while maintaining the re-
 5

quired network performance metrics to support delay-sensitive above, we can clearly see how the majority of the solutions
applications. This issue is further exacerbated in networks rely on simulations that do not completely model or capture
where the sensors follow a sleep-wake duty cycle to conserve various intricacies of a real-world wireless network. While this
energy. To overcome some of these challenges, [52] propose is a necessary and valuable step to design, evaluate, and refine
a cross-layer multi-path routing approach designed to support novel solutions, it often also ends up being the finish line for
duty-cycled WSN networks. Similar to [49], using the inter- several of these approaches due to the daunting challenges that
action between MAC and the network layer, this approach are involved in the maturation of the solutions. There have
also establishes multi-node disjoint paths with complementary been attempts to overcome this challenge and shortcomings
duty-cycling. The authors performed analytical study that by designing frameworks, and utilizing SDR-based testbed. In
proved the correctness of the proposed solutions. Simulations the next section, we discuss some of these attempts that have
were able to demonstrate up to 65% cumulative energy saving striven to bridge the gap.
when compared to the always-on multipath approach incurring
10% increase in response time delays.
B. Preliminary Testbed Evaluation
In [53], authors propose cross-layer clustering based routing
protocol for WSN to extend network lifetime. In this article, In several cases, the next logical step is to evaluate the fea-
the author select cluster head based on energy consumption, sibility and performance on a hardware-based testbed. The ad-
delay, throughput, security, distance and overhead. The authors vent of SDR has significantly nurtured efforts in this direction.
rely on MATLAB simulations to show how the proposed At the same time, there is a distinction in the maturity/utility
solution outperforms traditional approaches such as ant lion of solutions that have been implemented using a host PC con-
optimization approach and grouped grey wolf search optimiza- trolling the SDRs and ones efficiently ported to a standalone
tion. e-SDR. The key difference is in the rapidity of development on
In order to improve the energy efficiency of the IEEE computationally capable hardware in the first case as opposed
802.15.4 based network, the authors use a cross-layer approach to carefully optimized (often C/C++ or VHDL/Verilog) im-
to control the transmission power and minimize the broadcast plementation on embedded (resource-constrained) hardware.
of control packets [46]. The authors use information from the The host PC-based development approach saves time and
application layer, network layer, data-link layer, and physical resources to rule out impractical solutions before significant
layer to minimize the channel occupancy time by reducing time is spend in optimizing the implementation for final ma-
control packets (used for neighbor discovery). The proposed tured deployment. Therefore, based on the resources, the risk
approach is evaluated using NS2 to achieve approximately associated with novel cross-layer solutions either approaches
10% reduction in energy consumption while maintaining end- can be adopted to mature network control and management
to-end delays and comparable packet delivery ratios. solutions. To this end, several frameworks have been proposed
Another work that focuses on using cross-layer optimization for SDR to enable cross-layer optimized control [56]–[58].
for energy consumption is presented in [22]. In this work, There has also been work that employs SDR hardware but rely
the authors demonstrate how cross-layer optimization can be on emulation platforms [59] to perform evaluations as opposed
implemented to extend the network lifetime. To this end, ISA to over-the-air (OTA) experimentations. Such emulation-based
100.11 [54] and a Wireless highway addressable remote trans- approaches may provide more flexibility to perform larger
ducer (WiHART) [55] compatible sensor network is used in number of experiments in various topologies but still cannot
the context of petroleum refinery scenario. The authors demon- be substituted for OTA evaluations or field trials.
strate the utility of optimization that accounts for the fixed Several of the solutions discussed earlier have been success-
frame size constraint of current industrial wireless standards fully extended to preliminary hardware testbeds [41], [60]. In
in smaller network scenarios. Their NS3 simulations depicted most of these cases, SDRs like the universal software radio
longer network lifetimes than the traditional approaches that peripheral (USRP) are used in association with the host PC.
use minimum hop routing for the industrial wireless network. The authors extend their work of the proposed DRS algorithm
At the same time, they do caution by stating that even frame- [20] by implementing it on a five-node USRP testbed [41]. To
based optimization does experience a non-negligible failure accomplish this, the authors develop a cross-layer framework
rate when solving routes for large network sizes. they refer to as CrOss-layer Based testbed with Analysis Tool
A mathematical model for cross-layer protocol optimization (COmBAT) [58] that is implemented in Python to run on host
in SDR was proposed in [19]. The goal of the framework PC that controls the USRP SDRs. The authors compare the
is to provide tactical SDR with the flexibility to adapt its proposed solution with ROSA to demonstrate improvement in
objectives including minimizing energy consumption while both effective throughput and reliability of the network.
maintaining reliable packet delivery and latency constraints. Similarly, in [60] authors use four-node SDR testbed to
Though the proposed solution is designed for SDRs the demonstrate how cognitive channelization can be achieved by
solution is only implemented and evaluated in OMNET++. jointly optimizing the transmission power and the waveform
Their simulations used a number of mission-critical network channel of the secondary users. The experiments are conducted
scenarios to demonstrate enhanced performance where SDRs for both narrowband and wideband of primary users showing
effectively adapt to the dynamic environment. improvements in BER for both primary and secondary users.
As we survey through the breadth and width of cross- A resource allocation approach for CRN was designed and
layer solutions applied to achieve various objectives discussed implemented in [61] using two SDR-based testbeds of the
 6

ORCA federation [62]. The proposed distributed resource allo- in terms of metrics like throughput (due to low sampling
cation employs a Markov Random Field (MRF) framework to rate constrained by computations) and/or has been dependent
be deployed on secondary nodes of the CRN. The implemen- on external computational platforms. Due to these reasons,
tation leveraged GNU Radio to implement functionalities such even with the advances made in the field, there is no known
as spectrum sensing, collision detection, among others. The au- cross-layer optimized e-SDR solution built using commercial-
thors consider multi-channel CRNs and focus on the physical, off-the-shelf (COTS) e-SDR that, (i) can be deployed as a
data link layer (more specifically MAC), and network layer. standalone unit, (ii) makes cross-layer optimized distributed
The goal of any CRN is to ensure successful communication routing decisions, (iii) is ruggedized for outdoor deployment,
between secondary users in a dynamic environment without and (iv) can provide reliable and high throughput (up to
impacting the primary users. The secondary user SDRs are 11 Mbps) links over large distances (1 km for up to 5.5
designed to calculate an energy function based on the current Mbps). The difficulty of finding the right trade-off between
states of the neighbors. The objective of the secondary users is software design choices and performance along with achieving
to minimize the local energy function which in turn minimized standalone, fieldable hardware customization is a daunting
the interference. The authors claim that the secondary users task. This article provides such a solution for the first time
asymptotically converge to global optimal solution by updating showing that designing and deploying high Technology Readi-
their energy function through local sampling. The authors ness Level (TRL) cross-layer optimized solutions based on a
successfully implement the proposed solution on two SDR COTS low SWaP e-SDR is feasible.
testbeds, IRIS [63] and ORBIT [64]. The experiments ranged
from using up to 2 primary users and 9 secondary users. III. S YSTEM D ESIGN
They compare their solution with reinforcement learning- In this section, we first discuss some core design challenges
based channel allocation algorithms. Through their prelimi- that may be encountered during similar endeavors. Thereafter,
nary experimental evaluation, they were able to demonstrate we describe the system design of the cross-layer optimized
complete transparency towards primary users and demonstrate transceiver and how it has been executed.
performance in terms of collision percentage and the number
of required transmission slots for the secondary users of the
CRN. It is interesting to note that the achievable throughput A. Design Challenges
has not been discussed in this work which in most cases are 1) Choice of COTS SDR hardware: In recent years, there
throttled by lower sustainable sampling rates on the testbed. has been a surge in the number of SDRs available in the
Beyond relying on SDRs, well-defined commercial wireless market. While this has provided several tradeoff opportunities
protocols like WiFi and LoRa (for physical layer) have also for system design decisions, yet, determining the appropriate
been used along with microcomputing platforms to design choice is challenging due to the multi-dimensional tradeoff
cross-layer approaches [65]. In this case, a cross-layer ap- that exists with the decision making. At a high level, these
proach - distributed energy-efficient routing (SEEK) - was im- choices can be divided into some core characteristics portabil-
plemented on Raspberry Pi to maximize the network lifetime ity (size and weight), embedded computational capabilities,
using LoRa as the PHY. Using their proposed geographical RF frontend parameters (instantaneous bandwidth, tunable
routing protocol, SEEK, the authors were able to demonstrate frequency range, among others), and cost.
significant improvement in the network lifetime compared In our case, portability and embedded computational ca-
to greedy ad hoc routing approaches. In this form, it was pability were the two prime candidates. This meant ruling
highly restricted in throughput (due to LoRa) as a trade-off out all the SDRs that may not have sufficient embedded
for a longer transmission range. The survey of cross-layer computational resources. In this article, we refer to SDR with
approaches and their level of maturity has been summarized embedded processing capability (such as Field Programmable
in Table I. Gate Array (FPGA) and General Purpose Processors (GPP))
In all these examples, the achievable throughput is usually as e-SDR. Generally, the larger the SDR the higher the
low due to the lower sampling rate or rely on the significant computational resources but may end up consuming higher
computing power of the host PC. Either of these factors power and leading to a larger design. Hence, at this stage, the
renders such solutions highly restrictive in terms of utility, designer has to determine the best tradeoff that works for a
portability, fieldability, and often does not meet end-user given design and make sure the constraints from this point
requirements. In this article, we discuss how by customizing onwards is acceptable to the overall software architecture.
COTS low SWaP e-SDR with limited computing resources one 2) Limited computational resources: Embedded radio pro-
can efficiently design and deploy the first-known embedded gramming is a non-trivial task and the primary factor that
software-defined cross-layer optimized network. must be considered are the available computational resources.
In concluding this section, we have seen that the majority Accordingly, care must be taken to avoid redundant memory
of the cross-layer optimized works are limited to simulations read and write operations to reduce energy consumption and
and have not been successfully validated on a hardware latency of operation. From a protocol stack implementation
platform. This is a major hurdle and shortcoming of the perspective, the available accelerators are the CPU and FPGA
current state of research and development. In the recent past, housed in the Xilinx Zynq XC7Z010-2I SoC with 512 MB
some of this has been mitigated by preliminary hardware- of DDR3L RAM and 128 MB of QSPI Flash memory on the
based evaluations but has often provided limited performance radio platform. These impose constraints on the number and
 7

TABLE I: Summary table of cross-layer solutions and their maturity

Proposed Work Optimizing/ Simulation Preliminary Comments

Framework Testbed
Ding et al 2010 Throughput X Algorithm designed to perform joint routing, dynamic spectrum allocation, schedul-
[38] ing, and transmit power control for cognitive radio networks.
Ding et al 2013 Throughput X Algorithm designed to perform routing and code division channelization in a CDMA
[39] based cognitive radio networks.
Liu et al 2018 Throughput X Throughput-maximal cross-layer framework is proposed for DSRC-based vehicular
[48] network.
Jemili et al 2020 Throughput X Cross-layer multi-path routing approach that utilizes non-correlated and node-
[49] disjoint paths
Cheng et al 2013 Delay X Two cross-layer approaches are proposed to minimize end-to-end delay for multihop
[50] wireless network and evaluated in simulations.
She et al 2018 Reliability X The policies for packet dropping, power allocation, and bandwidth allocation are
[40] Latency optimized to minimize the transmit power under the QoS constraint.
Xlong et al 2011 Delay X A delay-aware joint flow control, routing, and scheduling algorithm is introduced
[51] for multihop network to maximize network utilization.
Xue et al 2013 Delay X Cross-layer approach that use virtual queue structures to guarantee finite buffer size
[47] or worst-case delay performance.
Jagannath et al Effective X Incorporates virtual queue length into the objective function of the cross-layer
2016 [20] Throughput algorithm to mitigate the last packet problem.
Wang et al 2006 Fairness X A dual-based algorithm and a penalty-based algorithm designed and evaluated to
[18] address the fair rate control problem in a multihop random access network.
Park et al 2008 Fairness X Authors propose a cross-layer feedback mechanism to improve fairness among
[42] sending and receiving WiFi stations. The solution is implemented and evaluated in
NS2.
Xu et al 2015 Fairness X Cross-layer fairness-driven approach to improve video delivery while remaining fair
[43] to the competing TCP flows. The solution is implemented and evaluated in NS2.
Herrmann et al Network Life- X Authors design cross-layer algorithm for extending lifetime of industrial network
2018 [22] time for petroleum refinery. The solution was evaluated using NS3.
Salameh et al Energy X Algorithm designed for CRN to minimize energy consumption while adhering to
2021 [45] consumption delay constraints and minimize impact on primary users. Algorithm outperforms
non-cross-layer approaches in MATLAB simulations.
Jemili et al 2021 Energy X Cross-layer multi-path routing approach designed to support duty-cycled WSN
[52] Consumption networks for energy efficiency while reducing impact on response time delays.
Lahane et al Energy X Cross-layer clustering based routing protocol that was shown to outperform tradi-
2021 [53] Consumption tional approaches like grouped grey wolf search optimization.
Al-Jemeli et al Energy X Cross-layer network operation model proposed for energy efficiency in IEEE
2015 [46] Consumption 802.15.4-based networks. The solution was evaluated using NS2.
Nosheed et al Energy X Mathematical model designed for SDR and performance evaluated in OMNET++.
2019 [19] efficiency
Herrmann et al Network Life- X Authors design cross-layer algorithm for extending lifetime of industrial network
2018 [22] time for petroleum refinery. The solution was evaluated using NS3.
Dai et al 2018 Energy X Authors aim to maximize energy efficiency by jointly optimizing modulation,
[44] efficiency transmit power, and route.
Shome et al Framework X Framework for implementation and evaluation of cross-layer approaches.
2015 [56]
Demirors et al Framework X Framework for implementing and evaluating cross-layer approaches.
2015 [57]
Jagannath et al Framework X Framework for implementing and evaluating cross-layer approaches.
2016 [58]
Soltani et al Not specified X Designed to integrate dynamic spectrum access, backpressure algorithm, and
2019 [59] network coding for multihop networking. Evaluation performed using emulations.
Jagannath et al Throughput X Throughput and reliability performance evaluated on 5-node USRP N210 SDR
2018 [41] testbed.
Sklivanitis et al SINR X BER performance evaluated on 4-node USRP N210 SDR testbed.
2015 [60]
Tsitseklis et al Avoid X Cross-layer framework designed for CRN and evaluated on 2-node SDR testbed
2020 [61] Collision named IRIS [63] and ORBIT [64].
SEEK 2019 [65] Network Life- X Algorithm considers link energy efficiency, backlog, forward progress, & residual
time energy. Solution is implemented on RPI & evaluated on 6-node testbed.
 8

size of the packet queues as well as storage of configuration timely information for the cross-layer decision engine.
and other header information on the radio.
Rather than going for a top-down approach whereby the B. Embedded Software Defined Radio Platform
protocol stack is designed and simulated with evaluations One of the key objectives of the work was to develop a
followed by hardware implementation, we chose a bottom-up modular, programmable, portable, handheld, battery-powered,
approach where these radio resource constraints are carefully standalone solution which should operate in harsh conditions
considered right from the inception of the protocol stack. for several hours. This implied that the foundation of the
These are evident from the energy consumption factor con- design needs to be a low SWaP e-SDR. The proposed solution
sidered in the routing, design of a segment size where the was implemented on a Epiq Solutions’ Sidekiq Z2 e-SDR [66]
packets are grouped into segments for outbound transmission, (Figure 1). It consists of an Analog Devices’ AD9364 RFIC,
and gathering Optimization Assisting Information (OAI) via Xilinx Zynq XC7Z010-2I system on chip (SoC), and the key
piggybacking. This way the control packet exchange won’t device specifications are provided in Table II.
occur per packet rather for a segment of packets. These
steps are adopted to reduce overhead which implicitly reduces TABLE II: Specifications of Z2
latency and energy consumption. Hence, the hardware con-
straints, as well as the target application requirements (long- Specs. Values
term unattended deployment), directed our design choices.
Frequency 70 MHz - 6 GHz
Most of the FPGA resources were utilized for the PHY
layer which is a pure IEEE802.11b standard. The PHY was Sample Rate Up to 61.44 MS/s
housed in the FPGA for reduced latency operation. Since the RAM Up to 512 MB of DDR3L
FPGA resources were nearly fully utilized for the 802.11b Flash memory 128 MB
implementation, none of the remaining upper layers could be Temperature rating −40C to +85 degC
offloaded to FPGA for acceleration. Consequently, the entire Processor Dual-core ARM Cortex A9
stack except the PHY/L1 was implemented on the Dual-core Size 30 x 51 x 5 mm
ARM Cortex A9 CPU of the SoC. The software architecture
Weight 8 grams
utilizes both the user space as well as kernel space. To reduce
repeated I/O operations and to parallelize the stack operations
to a full extent, we leveraged daemons executing from the Several hardware customizations were necessary to accom-
user space that hosted multi-threaded processes. A deep dive plish the objective of designing a fieldable transceiver using
into the software implementation is elaborated in section III-C. the COTS e-SDR. This includes adding a power amplifier,
The software architecture is designed and developed to execute filters for the frequency of interest, power supply system that is
agnostic to the underlying PHY/L1 layer, accordingly, we capable of supplying power from the battery during standalone
have followed a plug-and-play design ethos for the software remote operation but could also operate from a direct current
stack. Our implementation demonstrates this by interfacing (DC) power source when available. To aid the implementation
the software protocol stack executed fully from the ARM of various cross-layer routing techniques that use the location
processor with the 802.11b FPGA core. of nodes (such as geographical routing [65]), an embedded
3) Gathering optimization assisting information: Network GPS receiver was also included in the final design. The block
optimization can be generally divided into two categories, (i) diagram of the final transceiver and the ruggedized prototype
centralized and (ii) distributed. Centralized approaches often is shown in Figure 2.
provide globally optimal solutions but at the expense of large
overhead incurred in accumulating global information. This
also introduces delays and would suffer scalability issues for
larger network sizes. In contrast, distributed approaches reduce
the overhead and are in general scalable but may not converge
to globally optimal solutions. In either case, there is a need
to acquire information that is essential to execute cross-layer
optimization.
For ad hoc networks that are highly dynamic and need
to be scalable, distributed approaches are the desired choice.
This is also the case for our solutions. Our approach (which
will be discussed in detail in the next few sections) rely
on gathering OAI from immediate neighbors. To reduce the
overhead, this information can be appended to the control Fig. 1: Target baseline Z2 e-SDR
packets such as RTS, CTS, BEACON, etc. The frequency of
BEACON packets can be a function of network parameters
such as relative mobility of the nodes, rate of change of traffic, C. Cross-layer Software Architecture for Embedded System
among others. The essential takeaway is the need to carefully 1) Design Goals and Overall Architecture: The custom
design protocols to ensure efficient yet effective exchange of cross-layer protocol stack for the transceivers is implemented
 9

Main Antenna
TNC Connector GPS Antenna
RF SMA Connector
Filtering ARM CPU
GPS Receiver
Tx Power Outbound thread
Amp
K M Routing thread
System S Beacon thread General PHY/L1
e o

Netlink
Power Supply
o r d ZMQ thread Queue
Transmit Queue
Z2 API
c n u
k APP Receive thread Channel Access thread
e l
e l e X-n-APP Daemon Receive thread
PRC 148 t
Battery L2-3 Daemon
Ethernet PRC 148
with PoE Battery
Power Switch Zynq SoC
RJ45 12-28 V DC Input DC Power Input
RJ45 Jack
Fig. 3: Software Architecture.
Fig. 2: Transceiver Design

on the Zynq SoC of the Sidekiq Z2 platform. Specifically, the software. Essentially, designing the entire radio stack on the
PHY is an IEEE802.11b implemented purely on the FPGA SoC presents a daunting challenge owing to the memory, com-
subsystem while the upper layers are implemented in the putational, and latency constraints. An unorganized framework
C/C++ language on the embedded Linux operating system of could add overhead from unnecessary resource utilization
the Dual-core ARM Cortex A9 CPU of the SoC. consequently increasing the system latency.
As mentioned previously, our design principle is to develop The software architecture is broadly categorized into user
a plug-and-play software module of the protocol stack (except and kernel space with daemons running in the user-space
L1) to interface with any underlying PHY layer. The 802.11b as in Figure 3. We resort to daemons each of which hosts
FPGA core is a direct sequence spread spectrum (DSSS)- its own threads to attain a parallelized architecture that does
Complementary code keying (CCK) PHY with a 22MHz not perform redundant and unnecessary memory accesses. In
standard channel bandwidth which nearly fully utilized the other words, we can say the user-space hosts multi-threaded
logic cells and DSP blocks of the FPGA. The supported processes which interface with the kernel space. However,
data rates with DSSS are 1 Mbps and 2 Mbps while CCK there are certain design challenges with moving key functions
modulation supports 5.5 Mbps and 11 Mbps. The software to the kernel space. Kernel programming requires the most
stack is designed to be agnostic to the PHY layer and merely trusted operations as any bug or corruption may cause severe
interfaces with the PHY FPGA core over a C-level SKIQ80211 system crashes. Nevertheless, kernel space enjoys the benefits
application programming interface (API) provided by Epiq of low latency memory access operations. Furthermore, we
Solutions. The API provides all the necessary function calls also leverage the transport layer and IP headers supported by
to utilize the RF frontend. the Linux kernel. The software architecture in Figure 3 shows
We would like to reemphasize that the solution can be a socket which is a generalized custom socket architecture
adapted to use most if not all PHY layer standards as long for interfacing with external devices. The socket architecture
as the performance parameters (bit error rate, transmit power, is generalized to be able to reprogram and enable multiple
data rates, etc.) are communicated to the upper layers over interfacing options with a wide range of socket protocols such
an appropriate C/C++ level interface. Since the focus of this as ZMQ, UDP, etc.
work is the software implementation and evaluation of the The stack employs two daemons, namely; L2-3 and Cross-
upper layers (above PHY/L1), in the remainder of the section, layer & Application (X-n-APP) that interact with the kernel
we will detail the software architecture of the upper layers to module via Netlink sockets. Notice that the entire stack only
facilitate long-range mesh networking. has two daemons since it’s more efficient to use threads over
Our primary goal in building the software architecture was processes as they share the process’ resources. On the other
to maintain reconfigurability. The reconfigurability was en- hand, having more processes would require more computa-
forced by adopting a modular design framework with defined tional and memory resources. The kernel module handles the
functions for each module while being resource-efficient. We transport layer, packet encapsulation, and partial IP header
define reconfigurability as the ability to modify the protocol population prior to passing over to the L2-3 daemon. The L2-3
characteristics such as routing objective, specifics of the cross- daemon is the cross-layer L1, L2 (MAC), and L3 (Network)
layer information exchanged, etc. Accommodating a software- module which performs CSMA/CA-based medium access and
defined cross-layer protocol stack in an ARM CPU is a non- cross-layer routing. Another level of cross-layer interactions
trivial challenge. The software-defined architecture reaps ben- occurs when the L2-3 daemon acquires L1 information such
efits from the flexible software architecture which eases the fu- as the link reliability, data rate, etc., via the Sidekiq Z2’s
ture upgradability, customization (of factors like optimization SKIQ80211 API. Hence, the term cross-layer as it involves
objective among others), and maintenance of the transceiver interaction between L1, L2, and L3 to perform the optimized
 10

decision making. achieving maximum throughput is not the sole objective of

2) Cross-layer Routing Protocol Overview: The desired use SEEK algorithm, the differential backlog is just one parameter
case application of this work was to deploy an independent in our utility function. The effective progress made by a packet
energy-aware ad hoc network in large remote areas where can be represented as dis − d js . Choosing nodes that provide
traditional communication infrastructure for networks such as larger progress implies fewer hops to the sink node which
5G, WiFi, or LTE are not readily available. A prime example of in turn could lead to smaller energy consumption. Finally,
such an application are large test and evaluation (T&E) ranges to ensure uniform depletion of energy per node, we need to
j
that are usually set up in remote places, collecting data from consider the Er (residual energy) of potential next hops [70].
networks of remote installations, supporting rescue operations Therefore, we define our utility function as follows,
after a disaster where traditional communication infrastructure
is unavailable, among others. In many of these cases the core Relative Residual Energy
Differential Backlog
requirements of a desired routing protocol are as follows, j
!
max [∆Qi j , 0]
  
• The solution must be easy to deploy in ad hoc manner dis − d js Er
Ui j = ηi j j
, ∀ j ∈ NBi (2)
such that nodes can move in and out of the network qi dis E0
without effecting the overall network performance. Effective Forward Progress
• The solutions should be able to operate in a distributed
ηi j aims to improve the energy efficiency of the network
manner such that it can scale without incurring large
and can be replaced by reliability when using constant power
overhead.
and modulation. It is also interesting to note that the maximum
• Operates in an energy efficient manner to maximize
value of Ui j = ηi j when each of the three normalized terms
network lifetime since it is challenging to change batteries
is 1. This implies that each of the other terms penalizes the
regularly in a large remotely situated network.
utility function based on the instantaneous value. For example,
• Solutions should be able to adapt to changing network
a small differential backlog (qi − q j < qi ) will dampen the
traffic patterns and avoid congested paths by load bal- j
ancing. value of Ui j . Both dis − d js and Er will have similar effects
on Ui j .
Keeping the above requirements in mind, the diStributed
Energy Efficient bacKpressure (SEEK) routing algorithm [23], The objective of the network is to maximize the summation
[65] is leveraged in this work which utilizes the geographic of Ui j for all possible links L( i, j) in order to maximize the
information of nodes, differential queue backlog, residual overall energy efficiency of the network. This, in turn, will
battery energy, and link reliability to compute the optimal ensure reliable communication while maximizing the network
next hop. In this section, we will present a formal derivation lifetime (which is defined as the time when the first node
of our utility function (Ui j with respect to wireless link in the network depletes its energy leading to a network hole).
L( i, j)) between node i and node j and formulate the network The optimization problem is subject to residual battery energy,
optimization problem. queue backlog, bit error rate, and capacity constraints. This is
The utility function considers the following parameters formulated as Problem P1 shown below,
associated with potential next-hop; (i) proximity to destination,
(ii) differential queue backlog, (iii) residual battery energy,
P1 : Given : G (Nnet , E), G, Er , Q
(iv) energy efficiency of the link, and (v) the corresponding
link throughput. This information is gathered from traditional Find : NH∗ , T∗
control packets like the Beacon packets. The Beacon packets Maximize : ∑ ∑ Ui j (3)
will contain updated OAI. The energy efficiency of a given i∈Nnet j∈NBi
link can be expressed as follows [67], subject to :
ij
ij

ij
 Rb ≤ Ci j , ∀i ∈ Nnet , ∀ j ∈ NBi (4)
Rb 1 − eb Gi j ij ij
ηi j = = (1) eb > eb∗ , ∀i ∈ Nnet , ∀ j ∈ NBi (5)
Pi j Pi j
Eir > 0, ∀i ∈ Nnet (6)
where ηi j gives the measure of the number of bits success-
qi ≥ 0, ∀i ∈ Nnet (7)
fully transmitted over L( i, j) per Joule of transmission energy.
ij
Here, Gi j is the corresponding goodput measure and eb is the where the objective is to find the set of next-hop and
bit error rate with respect to L( i, j), Ti j is the transmission transmission strategy for all nodes in the network which can be
ij
strategy which includes choice of transmission data rate Rb represented as NH∗ = [NH∗i ] and T∗ = [T∗i j ] respectively, ∀i ∈
and transmit power Pi j . Nnet j ∈ NBi . In the above optimization problem P1 , G = [Gi j ],
Another key factor that needs to be considered in routing Er = [Eir ] and Q = [qi ], ∀i ∈ Nnet , ∀ j ∈ NBi denote the set of
is the differential queue backlog (∆Qi j = qi − q j ) with respect goodput measure, residual battery energy and queue backlogs
to the source node (i) and next-hop ( j) [20], [68], [69]. The respectively. The constraint 4 restricts the total amount of data
queue backlog at the destination node is considered to be rate in link (i, j) to be lower than or equal to the physical link
zero. Considering the queue backlog is necessary to mitigate capacity. Constraint 5 imposeS that any transmission should
congestion in the network and traditional backpressure algo- guarantee the required BER. Finally, constraints 6 and 7 ensure
rithms have been shown to be throughput optimal [68]. Since the residual energy and queue backlog of each node will not
 11

Beacon Header 13 if ( sourceBacklog ==0)

FC_1 FC_2 Duration_1 Duration_2 Addr_1 Addr_2 14 sourceBacklog =1;
1 Byte 1 Byte 1 Byte 1 Byte 6 Byte 6 Byte
15
Beacon Payload 16 double nextHopBacklog = ( double ) backlog ;
Message Type Size Node IP Node MAC GPS Location Utility # of Session 17 if ( isInfoPotentiallyStale () )
2 Byte 2 Byte 4 Byte 6 Byte 8 Byte 4 Byte 2 Bytes
Node # Battery Level Status Flags Channel Session Info
18 nextHopBacklog =0;
2 Bytes 2 Bytes 1 Byte 4 Bytes Variable 19
20 double minimumValue =0;
21 double diffBacklog =0;
Fig. 4: Beacon packet structure 22 if ( nextHopBacklog >0)
23 minimumValue =1/ nextHopBacklog ;
24 double BacklogTerm =( sourceBacklog - nextHopBacklog
have negative values. It can be seen that for solving the above )/ sourceBacklog ;
25 if ( nextHopBacklog < sourceBacklog )
optimization problem, nodes would require global knowledge 26 diffBacklog = BacklogTerm ;
of the network. This would require the centralized controller 27 else
to maintain largeR network tables storing node parameters 28 diffBacklog = std :: max ( BacklogTerm ,
minimumValue );
which will eventually flood with data as the the network 29
grows in size. Since the centralized optimization method is 30 // Compute residual battery energy
not a scalable solution, this motivates the need for a scalable 31 double nextHopBattery = ( double ) battery ;
32 double residualBattery = ( nextHopBattery ) /100;
distributed solution. We adopt SEEK which will operate in 33
a distributed fashion and enable each node to find the next- 34 // Compute utility term - Routing metric
hop based on the local information available to them. Each 35 double routeUtility = reliability * forwardProgress *
diffBacklog * residualBattery ;
node with a packet to transmit chooses an optimal next-hop 36
and transmission parameters such that it maximizes its own 37 return routeUtility ;}
local utility function. This can be considered as a divide-and- Listing 1: Routing Utility Computation Snippet - Battery
conquer approach to solving the optimization problem in a powered scenario
distributed manner. Accordingly, every source node (i) will
aim to maximize the utility function Ui j and select the optimal Here, line 35 shows the routing utility metric computation.
next-hop and transmission strategy as follows, This metric can be flexibly changed to include other route
optimization metrics. For example, if all the nodes are powered
[ j∗ , T∗i j ] = arg max Ui j , ∀ j ∈ NBi (8) by direct DC power supply and not on battery, then the routing
j
metric can be modified to address this scenario as in Listing
Each node will maintain a neighbor table with node param-
2.
eters of its neighbors and will update the table as needed based
on information from the beacon packets. The structure of the 1 // Compute utility term - Routing metric
2 double routeUtility = reliability * forwardProgress *
beacon packet is depicted in Figure 4 with self-explanatory diffBacklog ;
field names. The effectiveness of SEEK has been demonstrated
Listing 2: Routing Utility Metric Modification in scenario
in [65]. The emphasis of this work and contribution is not
when all nodes have DC power supply
SEEK but the design and feasibility of maturing similar
solutions for tactical applications. Due to the modular imple- Finally, for the benefit of the readers, we also provide
mentation, just by changing a few lines of code that defines Table III summarizing all the notations/symbols used in the
the utility function, one can reconfigure the stack to execute discussion above.
a new optimization objective. This flexibility in adapting the 3) L2-3 Daemon: The outbound DATA packets that are
routing metric is demonstrated in the code snippet in Listing handed over to the L2-3 daemon from the kernel module
1. are queued in the General Queue awaiting the best route
1 double calculateUtility ( neighborThroughputInfo & assignment. This is accomplished by the Outbound thread
sourceNode , neighborThroughputInfo & which continuously listens for incoming packets from the
destinationNode )
2 { kernel module. The SEEK thread continuously performs route
3 // Compute reliability term computation and assignment for the outbound packets in the
4 double reliability = reliabilityMean () ; General Queue. Following route assignment, the packets will
5
6 // Compute forward progress term await their transmission opportunity in the Transmit Queue. A
7 double distanceToTarget = haversine_distance ( third thread - Channel Access - performs CSMA/CA awaiting
sourceNode . latitude , sourceNode . longitude , a clear-to-send (CTS) from its intended next-hop/destination to
destinationNode . latitude , destinationNode .
longitude ); dequeue the segment (set of DATA packets) from the Transmit
8 double distanceToTargetFromNextHop = Queue. The segment for which the CTS was received will
haversine_distance ( latitude , longitude , be forwarded to the FPGA via the L1 API for over-the-air
destinationNode . latitude , destinationNode .
longitude ); transmission. It must be noted that if the intended next-hop for
9 double forwardProgress =(( distanceToTarget - a segment is unresponsive, the segment will be returned back
distanceToTargetFromNextHop )/ distanceToTarget ); to the General Queue for rerouting. Additionally, a Receive
10
11 // Compute buffer backlog term thread continuously monitors for incoming packets from the
12 double sourceBacklog =( double ) sourceNode . backlog ; FPGA, processes, and responds (such as with CTS or ACK)
 12

TABLE III: Summary of Notations layers for outbound transmission where the remaining fields
such as residual battery and current buffer backlog are updated.
Symbols What it represents Here, we note that, unlike the DATA, the beacon packet is
L(i, j) Wireless link between node i and node j treated as a control packet and is not queued in the general or
transmit queues of the L2-3 daemon rather it is directly sent
Ui j Cross-layer utility function representing the
to the L1 FPGA for transmission.
efficacy of wireless link L(i, j)
The X-n-APP daemon also functions as a generalized appli-
ηi j Number of bits successfully transmitted over
cation daemon for direct interfacing with an application such
L (i, j) per Joule of transmission energy as a GUI. Both ZMQ and APP receive threads are designed to
Gi j Goodput of wireless link L(i, j) support such applications and hence can be customized based
ij
eb Bit error rate of wireless link L(i, j) on the application at hand. Since the use case of this particular
ij
eb∗ Minimum acceptable bit error rate transceiver is remote deployment in test and evaluation ranges,
ij a GUI application is developed to interface with the transceiver
Rb Data rate chosen for L(i, j)
Pi j Transmit power chosen for L(i, j) at a central command and control location. We emphasize that
the remote monitoring and configuration (parameters like data
Ti j Transmission strategy comprising of data rate
rates, frequency, among others.) capability enabled by the GUI
and transmit power
will ease the operator load by alleviating the need to physically
qi Backlog of node i travel to the deployed locations. The functioning and design
∆Qi j Differential Backlog of the GUI has been described in detail in Section III-D.
dis Distance between node i and destination s We further reemphasize that the software-defined stack is
j
Er Residual energy of node j designed to be reconfigurable to modify the cross-layer routing
j as the requirements evolve in the future or depending on the
E0 Total initial energy of node j
NBi Set of neighbors of node i desired network application. It is noteworthy that the utility
function discussed above is representative of one such example
P1 Optimization problem 1
of the algorithm where the network application desires energy-
G (N, E) Graph with nodes N and edges E aware routing. This specific choice was application-specific.
Nnet Set of all nodes in the wireless network The broad impact of this work is the reconfigurable nature of
E Set of all edges between nodes in the network the software-defined stack such that it keeps evolving to meet
Er Set of residual energies of all nodes in the the future requirements rapidly.
network
G Set of all goodput measures D. Graphical User Interface Application
Q Set of all node backlogs
In this article, we provide an example of how an application
NH∗ Set of all optimal next hop for the given time
such as a GUI can be configured for the network using the
step
proposed software architecture. As described in the previous
T∗ Set of all optimal transmission paramters for section, remote network (re-)configuration and monitoring are
the given time step achieved with the GUI shown in Figure 5 coupled with a
NHi Set of all neighbors of node i central node (referred to as gateway node). The gateway
Ci j Maximum channel capacity node is the only node that directly communicates with the
GUI (which runs on a host PC) via Ethernet. The GUI
issues command packets to the gateway node which are then
transmitted to the other nodes over the air. Multiple packet
accordingly based on received packet type. A supplementary types are designed to issue specific commands. These packets
auxiliary thread also tracks DATA timeouts. are divided by their purpose and target (i.e. entire network
4) X-n-APP Deamon: The X-n-APP daemon handles two or just a particular node). Essentially, each of the new packet
tasks; (i) preparing the beacon packet with the necessary infor- types is associated with a command and can have a unicast or
mation to be shared with immediate neighbors for distributed broadcast variant depending on the intended destination(s) of
optimization, and (ii) handling packets for any application such the packet.
as a graphical user interface (GUI) in this case. The beacon The GUI uses a build system of Babel and Webpack. This
packets as well as the GUI requires location information to ensures the GUI has a small footprint and is both faster
reflect the most recent coordinates on the GUI. We choose to develop on and faster to run as an end-user. The GUI
an efficient approach and access the GPS module from only is split into two processes: Main and Renderer. The Main
one location - X-n-APP daemon - in the stack to avoid GPS process handles the communication to the network (through
pinging from multiple software locations. This retrieved GPS the gateway node). The Renderer process handles user input
location information is subsequently populated in the Beacon and displaying output information that is used to interact with
packet. The GPS location extraction as well as beacon packet the GUI. When the user wants to issue a command to the
construction is carried out by the Beacon thread. The Beacon network, they are required to enter the value they would like
thread periodically sends these beacon packets to the lower to change and click the button labeled Update Parameter. Once
 13

Fig. 5: Graphical User Interface (GUI) Application

the Update Parameter button is clicked, the Renderer process the receiving node including the node’s IP address and GPS
of the GUI will create a control packet and send it to the location. Once the responses reach the gateway node, they are
Main process of the GUI. Upon receiving the control packet, sent back to the GUI via the dealer/dealer ZMQ socket. The
the Main process will send it to the gateway node on a ZMQ GUI will use this information to display the node’s status. If
Dealer/Dealer socket. The GUI is constantly listening to the after a set duration, a node’s response does not reach the GUI,
Dealer socket for any packet sent from the network. the status of the node will be set to Disconnected in the GUI.
After receiving the control packet, the gateway node will An example of the list of nodes in the network displaying its
extract the command and target fields of the packet and use status can be seen below in Figure 6. The battery bar being
them to determine what type of packet - broadcast or unicast grey, indicates DC power in addition to battery power. No
- needs to be sent. For example, the command could be Get battery bar indicates only DC power, and a green bar is only
Status and the target could be all the nodes in the network battery power. The list of devices also includes action buttons
which will cause the gateway node to respond by issuing the that are Get Network Status, Reset Network, and Announce
Get Status command for all nodes using a broadcast packet. Gateway respectively. These are displayed in the upper right
A broadcast packet is broadcasted for all nodes in range of the table.
of the gateway node and then rebroadcast by each node that Figure 7 shows the Message List tab of the sidebar with the
receives it for the first time. Upon receiving a broadcast packet, Success section toggled. It is used to view the status of packets
a node will process the packet, strip its header, and send the sent from the GUI. The Sent section displays packets sent
packet payload up to user-space for packet handling. A unicast from the GUI. The Pending section displays packets that are
packet is sent to a target node on the route determined by yet to reach the gateway node. The Success section is where
the SEEK routing daemon using the target IP field from the packets that have received a response appear. All indicates
GUI’s command packet. The reset command is utilized with the packet sent was a broadcast packet intended for every
two different packet types, Reset Node and Reset Network. node in the network. An Announce Gateway for All will be
The Reset Network packet is a broadcast packet and is used listed under success when the gateway successfully receives a
to force each node on the network to restart the protocol response from any node in the network. An IP address, such
stack. Restarting involving safe closing and exiting of all active as "10.241.37.131“, indicates the packet was unicast. A Get
daemons and starting the software stack again. The Reset Node Status for "10.241.37.131“ will be listed under Success once an
packet is a unicast packet that forces a specific target node to acknowledgement from the node at IP address "10.241.37.131“
restart. is received at the Gateway. Figure 7 also shows the Config tab
There are two packets that utilize the Get Status command, of the sidebar in the GUI and is used to execute commands
Request Network Status and Request Node Status. A Request on a selected node. Figure 5 shows the map view in the GUI
Network Status packet is a broadcast packet that is sent upon which reflects the network topology. The gateway node is
the gateway receiving a Get Status command from the GUI displayed as a blue dot to make it evident to the user.
that targets all nodes on the STRAIN network. The Request Since the gateway node is the only node that can directly
Node Status packet is a unicast packet that is sent upon the communicate with the GUI, the gateway must recognize every
gateway node receiving a Get Status command for a target new node added to the network. Upon startup, each new node
node. Upon receiving one of these packets from the gateway, broadcasts a New Node Announcement packet to neighboring
the receiving node will send a Response to Gateway packet as nodes requesting the gateway node IP. This request is broad-
an acknowledgment to the gateway. cast until a neighboring node responds with the requested IP
A Response to Gateway packet will include OAI about in a New Node Announcement Response packet. As a result,
 14

Network Parameter or Set Node Parameter packet. It is only

created and sent after the change occurs. Summary of remote
control and monitoring capabilities are provided in Table IV

IV. O UTDOOR E XPERIMENTS AND R ESULTS

As part of the feasibility analysis, we performed exten-
sive outdoor experiments to evaluate the performance of the
Fig. 6: GUI’s device Status Page implemented software-defined protocol stack in a realistic
environment. Since the proposed solution is implemented on
an e-SDR, it can be extended to any frequency of interest sup-
ported by the target hardware. To accomplish the experiments,
we received a temporary experimental license from Federal
Communications Commission (FCC) to utilize the 430 MHz
frequency within a geographical area. The novelty and utility
of the SEEK algorithm itself have been established in previous
work [65], [71]. Hence here the focus is to compare various
settings and the impact on overall performance (through-
put and reliability) of implementing cross-layer optimized
algorithms on GPP supported with FPGA-based PHY layer
for outdoor deployment. To the best of our knowledge, this
is the first time such a comprehensive evaluation has been
Fig. 7: Sidebar of the GUI undertaken using deployment-ready cross-layer optimized e-
SDRs. The default parameters are listed in the Table V unless
otherwise specified.
the new node and the neighboring nodes update the node list
and neighbor lists. Once the new node has the gateway IP, it
repeatedly sends a Hello Gateway packet until the new node is A. Transmission Range Evaluation
acknowledged by the gateway. Once the gateway receives this The target deployment scenario for the device is remote test
Hello Gateway packet, the gateway informs the GUI of the and evaluation sites, hence, the design goal for transmission
new node and sends a Gateway New Node Acknowledgement range was to achieve up to 1 km. The setup of the transmitter
completing the addition of a new node to the network. and the key testing locations from our range testing experi-
A Set Network Parameter packet is broadcast to all nodes ments is shown in Figure 8. Since the authorization to transmit
in range of the gateway node. The three commands currently was only for a limited geographical area, we redesigned the
available with Set Network Parameter are Update Data Rate, MAC protocol to ensure the receiving node does not transmit
Update Tx Power, and Update Frequency commands which any control packet and just acts as a receiver for the purpose
are sent throughout the network starting from the gateway of range testing. This ensured that the receiver that needs to
and being rebroadcast from the nodes until every node has be outside the geographical locations does not transmit and
received the command. The command Update Data Rate has just acts as a passive receiver to evaluate the reliability of the
four values associated with that can be sent through the link from the transmitter. The rest of the software protocol
network. It changes current data rate of every node in the stack was the same as the architecture discussed in Section
network to the supported values. In this work, it was set III. For each reported value in the Table VI, the results were
to one of the four values, which are 1 Mbps, 2 Mbps, 5.5 averaged over 10,000 packets transmitted in each run. We
Mbps, and 11 Mbps. The command Update Tx Power sets the define reliability as the percentage of packets received with
attenuation for every node in the network based on the frontend respect to packets sent.
supported values. Similarly, the Update Frequency command Table VI shows that the performance was consistently good
will set the frequency of the nodes to the specified value. A for both 2 Mbps and 5.5 Mbps up to 1019 m but the perfor-
Set Node Parameter packet is unicast from the gateway node mance of 11 Mbps degraded sooner. Due to the FCC license
to a targeted node in the network. The commands currently restrictions and urban environment, the next feasible point was
available with Set Node Parameter are Update Data Rate and at 1942 m from the transmitter at which point no packets
Update Tx Power. The Update Data Rate and the Update Tx were received. The performance degradation of 11 Mbps
Power command have the same possible values as above for was attributed to sensitivity to lower signal-to-noise-ratio and
Set Network Parameter. multipath propagation effects. This was further substantiated
An Acknowledge Gateway packet is transmitted by non- since the performance did improve when we tested a smaller
gateway nodes after the nodes receive a Set Network Parameter payload size. For example, reducing the payload size to 100
or a Set Node Parameter packet. The Acknowledge Gateway Bytes at a distance of 1019 m increased the reliability from
packet is a unicast packet from the non-gateway node to the 13.9% to 30%. Further, we also analyzed the Received Signal
destination (gateway) node. Its purpose is to acknowledge the Strength Indicator (RSSI) values which provided us insight
gateway node after making the requested change with Set that the performance of the physical layer was deteriorating at
 15

TABLE IV: Summary of GUI Commands

Application Command Description

Get Node Status Gets the status of the selected node
Get Network Status Gets the status of the entire networks
Announce Gateway Announces the gateway node’s information to all nodes connected on the network
Set Data Rate: Apply to Node Sets the data rate of the selected node.
Set Frequency Sets the frequency of the network
Set Transmit Attenuation Sets the transmit power of the a target node or the entire network
Reset Node Forces a selected node to restart
Resent Node Forces the entire network to restart

TABLE V: Parameters Of Outdoor evaluation TABLE VI: Link reliability for varying transmission ranges

Parameters Values Data Rate Distance Reliability

Data rates 1, 2, 5.5, 11 Mbps 495.2 m 99.9%
Transmit Power 1-3.5 W (based on attenuation) 771.2 m 99.77%
2 Mbps
Center Frequency 430 MHz 1019 m 98.08%
Bandwidth 22 MHz 495.2 m 99.62%
Payload Size 1000 Bytes 771.2 m 96.03%
5.5 Mbps
Segment Size 32 Packets 1019 m 97.16%
495.2 m 85.28%
771.2 m 31.56%
TX Location 11 Mbps
1019 m 13.9%
X=0

this, we perform outdoor peer-to-peer testing for varying data

RX Location 1 rates and payload sizes to analyze the impact. The outdoor
X=495.2 m locations of the transmitter and receiver are depicted in Figure
9.
Transmitter Set up at TX Location, X=0
First, in Figure 10, we study the reliability at the different
RX Location 2
data rates with and without automatic repeat request (ARQ)
X=771.2 m
enabled at the MAC layer. It can be clearly seen that the
transceiver has high reliability close to 100% for 1, 2, and 5.5
RX Location 3
Mbps even without ARQ enabled. At 11 Mbps the reliability
X=1019 m
decreases to 92% without ARQ and returns to 100% with
Fig. 8: Range testing setup and map-view of the locations. ARQ by trading off throughput. Figure 11 depicts how the
peer-to-peer throughput varies at different PHY data rates 1,
2, 5.5, 11 Mbps. The throughput in this case only considered
RSSI of ∼ 15 dB or higher than that observed during wired
the payload of the packets and does not include the headers
in-laboratory experiments. This finding revealed there is room
or the control packets. The throughput considered here is
for improvement in sensitivity (∼ 15 dB) in future iterations of
also referred to as goodput in some literature. The figure
the FPGA PHY. Overall, the experimentation was successful
also depicts the normalized throughput which is defined
as we hit our target distance of 1 km even at 5.5 Mbps.
as the ratio between throughput and data rate. The design
choice provides solid performance at 1 and 2 Mbps especially
B. Peer-to-Peer Experiments considering it depicts the goodput achieved. The impact of
As described earlier, the decision to implement the PHY on the overhead becomes visible as the data rate increases to 5.5
FPGA and the rest of the protocol stack was to achieve the best and 11 Mbps. This hypothesis is further substantiated in our
combination of high data rates and the required flexibility to experiments that show an increase in normalized throughput
implement cross-layer routing protocols. Yet, it is important to with the increase in the payload as shown in Figure 12. The
study the delays and associated overhead it takes to make the normalized throughput increased by up to 39% for 11 Mbps
necessary calls between the GPP and FPGA to support packet when the payload was increased from 1000 Bytes to 3000
transmission and how it impacts the data rates. To accomplish Bytes.
 16

5
102 Without ARQ Without ARQ
TX Normalized Throughput With ARQ
With ARQ
1
4 0.83
100 99.9 100.0 99.9 100.0 99.9 100.0 100.0 0.75
3.57
0.55

Throughput (Mbps)
0.5
98 0.32 3.05 3.02
3

Reliability
2.82
185 m
96 0
1 2 5.5 11
2
94 1.50 1.49

92.7
1 0.83 0.83
RX 92

90 0
1 2 5.5 11 1 2 5.5 11
Data rates (Mbps) Data rates (Mbps)
Fig. 9: Peer-to-Peer
Topology Fig. 10: Reliability vs data rates Fig. 11: Throughput vs data rates

9 to node 6. The experiments were conducted with and without

Improvement in Payload=1000 Bytes
8 Normalized Throughput Payload=2000 Bytes
ARQ. To ensure a line network in the limited geographical
12.92 % Payload=3000 Bytes area authorized for experiments by the FCC, we had to set
7 1 19.73 % neighbor tables in software. This also implied that all nodes
28.94 %
Throughput (Mbps)

6 will cause interference to each other which also serves as a

0.5 39.01 %
good test for ad hoc channel access capabilities/performance.
4.95
5 The results followed exactly the hypothesized pattern in the
4.56

4
0
1 2 5.5 11
3.93 case of normalized throughput. In the single-hop a normalized
3.51 3.56
3.04
throughput of 0.83 and all the other measures for hops 2, 3,
3 4, 5 followed close to 1/2, 1/3, 1/4, and 1/5 of the normalized
2 1.701.80
throughput of the single-hop case as shown in Figure 14. This
1.50
implied that the nodes were utilizing (sharing) the spectrum
1 0.820.910.93
efficiently without too much loss due to collision. As shown
0 in Figure 15, the reliability also seemed to confer with this
1 2 5.5 11 finding as in most cases reliability was over 95%. The lowest
Data rates (Mbps) reliability was still 93.9% for the 5-hop case even when the
Fig. 12: Throughput for varying payload ARQ was disabled. All the reliability was back to 100%
when ARQ was enabled. There was only a marginal drop in
normalized throughput when ARQ was enabled which was
mainly attributed to the link being very reliable even without
ARQ leading to only a few retransmissions even when ARQ
was enabled.
Node 6
Node 5
D. 5-Node Network Experiment - Dynamic Routing
Node 4
Node 3 For timely evaluation, we designed a specific experiment
Node 2 such that we can determine the effectiveness of the implemen-
Node 1 tation of SEEK algorithm to dynamic changes in the network.
To accomplish this, we consider a topology shown in Figure
16. The source S1 continuously transmits packets destined for
the gateway node GW. S1 chooses the appropriate relay among
R1, R2, and R3 based on SEEK algorithm. We plot the average
packets received from each relay node every 10 s interval. The
Fig. 13: 6-node Line Network Topology values are averaged using a moving window of 60 s duration
to get smoother curves.
In the first part of Figure 18 (blue shade) depicts a scenario
C. 6-Node Line Network where the relay nodes (i.e. nodes with possible forward
In the next set of experiments, we examined the impact progress for S1) have approximately uniform parameters. As
of the number of hops on normalized throughput and the expected, in this scenario, the traffic is evenly distributed
reliability of the network. The line network topology is shown between the three relays. Next, we significantly increase the
in Figure 13. In this topology only the nodes adjacent to each backlog of the R1 to emulate a congestion scenario at a relay
other are neighbors. Hence five hops are required from node 1 node. As it can be seen (yellow shade), S1 learns to avoid
 17

0.9
0.83 0.83 Without ARQ 102 Without ARQ
0.8 With ARQ With ARQ

100.0100.0 100.0 100.0 100.0 100.0

0.7 100
Normalized Throughput

98.8 98.6
0.6
98

Reliability
0.5 96.9

0.40 0.40 96
0.4

0.3 0.26 0.26 93.9

94
0.20 0.20
0.2
0.15 0.14
92
0.1

0 90
1 2 3 4 5 1 2 3 4 5
Number of hops Number of hops

Fig. 14: Line Network Throughput Analysis Fig. 15: Line Network Reliability Analysis

Source 3

Relays
R3 Source 1
R2
R3 New Location
GW R1 Gateway
Gateway/ S1 Source 4
Source 2
Destination Source

Fig. 16: 5-node Outdoor Topology Fig. 17: 10-node Outdoor Topology

15 0.7
Relay R1 R1 Backlogged Node R1 Backlog
R2 Battery drops R3 location change
cleared
Relay R2
Relay R3
0.6
Total
Normalized Throughput

0.5
10
Avg. No. Packs

0.4

0.3

5
0.2
Pak/s=10
Pak/s=20
0.1 Pak/s=40
Pak/s=80

0 0
2 4 6 8 10 12 14 16 18 20 22 1 2 4
Time(min) Number of sessions

Fig. 18: 5-node Experiment (dynamic routing) Fig. 19: 10-node Experiment

R1 and routes the packets through the other two options, R2 the evaluation, the residual energy of R2 is rapidly dropped
and R3. At the beginning of the experiment, R2 was set up by removing the DC power supply and the residual energy
with a battery with low residual energy but was connected in the battery was close to 10%. As it can be seen, S1
to a DC power supply. When a node is connected to a DC recognizes this in the next part (yellow shade) and prefers R3
power supply, it is equivalent to having the battery completely so that lifetime of the network can be extended. We remind
charged with full residual energy. Now in the next part of the readers that the lifetime of the network is defined as
 18

the duration of network operation till the first node in the V. C ONCLUSION AND F UTURE D IRECTION
network is completely depleted of residual energy. Therefore, This article introduces the first-known fieldable cross-layer
the nodes use the utility function described earlier in Section optimized, embedded software-defined radio (e-SDR) that
III trying to jointly consider multiple parameters including has been designed, developed, and matured to provide high
residual energy. Therefore, one goal is to distribute traffic throughput and reliability by implementing the protocol stack
over multiple paths so that nodes in the network can share (except the physical layer) on an embedded ARM processor.
the load based on available energy resources to extend the To emphasize the contribution of the article, we survey the
lifetime of the network. This is the exact intuitive behavior literature to highlight that the majority of cross-layer optimized
observed here in this region of Figure 18. Thereafter, in the techniques have been either limited to simulations or extended
next part, we now move the R3 node behind the source S1 as to hardware testbeds. We point out some of the key hurdles that
shown in Figure 16 such that it is no more a feasible next-hop might have led to the halt in the maturity of these solutions.
that provides forward progress. This information is quickly The survey also demonstrates the large scale of applicable do-
realized by the source S1 through the beacon packets as in mains for cross-layer optimization if the solutions are designed
all the previous cases. In this situation, with R1 congested, and matured to be deployed in real-life applications.
R3 is no more the possible choice for forward progress, S1 First, to overcome the hardware-related challenges, we
must return to R2 with a lower battery since that is the only discuss how a COTS e-SDR can be customized and re-
possible choice to relay the packets to the destination (GW). configured for deploying cross-layer protocol stack. We then
Hence in the red region of Figure 18, it can be seen that provide a detailed discussion on the software architecture
the traffic through R2 picks up while the traffic through R3 and design choices that ensured the feasibility of such an
drops. Another interesting observation is that when the R3 endeavor. Ensuring the right balance between modular and
initially changes location it still has some packets remaining flexible architecture along with efficiency is the key takeaway
for the GW, and GW is still within its transmission range. for this discussion. Most importantly, we have demonstrated
Accordingly, after the initial drop in traffic, it still is able through extensive outdoor field experiments the capability,
to get those packets to the GW which can also be seen range, and dynamic routing under varying network conditions.
as a spike in the total traffic momentarily to overcome the In the future, cross-layer optimized networking can be a
dip it experienced a few moments prior. In the final section key enabler for several applications in IoT, wireless sensor
of the Figure 18, the backlog of R1 is cleared. Now, in networks, 5G, and beyond. For example, device-to-device
this case, the S1 realizes that it does not have to use the communication can be enhanced by cross-layer optimization
R2 with lower residual energy anymore and can share the that could be implemented on resource constrained edge
traffic load through R1. As a result, it can be seen that platfroms. Additionally, several of the tactical applications
now traffic through R1 increases while the traffic through R2 rely on ad hoc networking to provide the much needed
reduces. It is also interesting to see that even with all these flexibility during tactical operations. We hope this successful
changes, the total packets/s at the GW (represented at the demonstration of cross-layer optimized e-SDR will provide
"total" legend in Figure 18) remains very stable demonstrating directions for future cross-layer optimized solutions to benefit
the rapid adaptability of the cross-layer optimized nodes. In tactical and commercial applications.
this experiment, we have demonstrated the gamut of routing
decisions the nodes can make in a distributed manner by
just gathering information from their immediate neighbors ACKNOWLEDGMENT AND D ISCLAIMER
demonstrating the effectiveness of the cross-layer optimized The authors would like to thank John Orlando, Jeff Porter
routing using an intuitively representative experiment. of Epiq Solutions for their support, Raymond Shaw of Spec-
trum Bullpen for help with Spectrum Supportability Risk
E. 10-Node Network Experiments - Network Capacity Assessments, and Dan O’ Connor of ANDRO Computational
In this experiment, we evaluate how the network handles Solutions for his help during the outdoor testing.
the increasing number of sessions and source (packet gen- (a) Contractor acknowledges the Government’s support in
eration) rate. The 10-node topology is shown in Figure 17. the publication of this paper. This material is based upon
The experiments were conducted at 1 Mbps of data rate, the work supported by the US Army Contract No. W15P7T-
source rate was varied from [10, 20, 40, 80] packets/s, and the 20-C-0006. (b) Any opinions, findings, and conclusions or
number of sessions (independent sources generating traffic to recommendation expressed in this material are those of the
the gateway node) was set to 1, 2 and 4. This is motivated author(s) and do not necessarily reflect the views of the US
by a typical use case for our intended deployment scenario. Army.
As expected, Figure 19 demonstrates that the normalized
throughput increases and gets saturated both when the source R EFERENCES
rate increases and the number of sessions increase. There is
[1] M. Katz, M. Matinmikko-Blue, and M. Latva-Aho, “6genesis flagship
no drop in performance even when there are multiple sessions program: Building the bridges towards 6g-enabled wireless smart soci-
(4) with high source rates (80) and the normalized throughput ety and ecosystem,” in Proc. of IEEE Latin-American Conference on
is maintained at the saturated value (∼ 0.6). This shows the Communications (LATINCOM), 2018, pp. 1–9.
[2] Y. Zhao, J. Zhao, W. Zhai, S. Sun, D. Niyato, and K.-Y. Lam, “A survey
efficiency of the network in handling multiple traffics and of 6g wireless communications: Emerging technologies,” ArXiv, vol.
session rates as expected even when the traffic increases. abs/2004.08549, 2020.
 19

[3] A. Jagannath, J. Jagannath, and T. Melodia, “Redefining wireless Mission,” in Proc. of IEEE Cognitive Communication for Aerospace
communication for 6G: Signal processing meets deep learning,” IEEE Applications (CCAA) Workshop, Cleveland, OH, USA, June 2019.
Transactions on Artificial Intelligence, 2021. [27] Y. Zhou, H. Yang, Y.-H. Hu, and S.-Y. Kung, “Cross-layer network
[4] E. Calvanese Strinati, S. Barbarossa, J. L. Gonzalez-Jimenez, D. Ktenas, lifetime maximization in underwater wireless sensor networks,” IEEE
N. Cassiau, L. Maret, and C. Dehos, “6g: The next frontier: From Systems Journal, vol. 14, no. 1, pp. 220–231, 2020.
holographic messaging to artificial intelligence using subterahertz and [28] D. Pompili and I. F. Akyildiz, “A multimedia cross-layer protocol for
visible light communication,” IEEE Vehicular Technology Magazine, underwater acoustic sensor networks,” IEEE Transactions on Wireless
vol. 14, no. 3, pp. 42–50, 2019. Communications, vol. 9, no. 9, pp. 2924–2933, 2010.
[5] G. Berardinelli, N. H. Mahmood, I. Rodriguez, and P. Mogensen, [29] L.-C. Kuo and T. Melodia, “Cross-layer routing on mimo-ofdm under-
“Beyond 5g wireless irt for industry 4.0: Design principles and spectrum water acoustic links,” in 2012 9th Annual IEEE Communications Society
aspects,” in Proc. of IEEE Globecom Workshops, 2018, pp. 1–6. Conference on Sensor, Mesh and Ad Hoc Communications and Networks
[6] S. Dang, O. Amin, B. Shihada, and M.-S. Alouini, “What should 6g (SECON). IEEE, 2012, pp. 227–235.
be?” Nature Electronics, vol. 3, pp. 20–29, 2019. [30] J. Liu, M. Yu, X. Wang, Y. Liu, X. Wei, and J. Cui, “Recrp: An un-
[7] W. Saad, M. Bennis, and M. Chen, “A vision of 6g wireless systems: derwater reliable energy-efficient cross-layer routing protocol,” Sensors,
Applications, trends, technologies, and open research problems,” IEEE vol. 18, no. 12, p. 4148, 2018.
Network, vol. 34, no. 3, pp. 134–142, 2020. [31] M. C. Vuran and I. F. Akyildiz, “Cross-layer packet size optimization
[8] L. Bonati, M. Polese, S. D’Oro, S. Basagni, and T. Melodia, “Open, for wireless terrestrial, underwater, and underground sensor networks,”
Programmable, and Virtualized 5G Networks: State-of-the-Art and the in IEEE INFOCOM 2008-The 27th Conference on Computer Commu-
Road Ahead,” Computer Networks, vol. 182, pp. 1–28, December 2020. nications. IEEE, 2008, pp. 226–230.
[9] B. Fu, Y. Xiao, H. Deng, and H. Zeng, “A survey of cross-layer designs [32] J. M. Jornet, M. Stojanovic, and M. Zorzi, “On joint frequency and
in wireless networks,” IEEE Communications Surveys Tutorials, vol. 16, power allocation in a cross-layer protocol for underwater acoustic
no. 1, pp. 110–126, 2014. networks,” IEEE Journal of Oceanic engineering, vol. 35, no. 4, pp.
[10] F. Foukalas, V. Gazis, and N. Alonistioti, “Cross-layer design proposals 936–947, 2010.
for wireless mobile networks: A survey and taxonomy,” Commun. [33] N. Cen, J. Jagannath, S. Moretti, Z. Guan, and T. Melodia, “LANET:
Surveys Tuts., vol. 10, no. 1, pp. 70–85, Jan 2008. Visible-Light Ad Hoc Networks,” Ad Hoc Networks, vol. 84, 2019.
[11] V. Srivastava and M. Motani, “Cross-layer design: a survey and the road [34] M. S. Demir and M. Uysal, “A cross-layer design for dynamic resource
ahead,” IEEE Comm. Magazine, vol. 43, no. 12, pp. 112–119, Dec 2005. management of vlc networks,” IEEE Transactions on Communications,
[12] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, “A survey on cross- vol. 69, no. 3, pp. 1858–1867, 2021.
layer quality-of-service approaches in wsns for delay and reliability- [35] J. Jagannath and T. Melodia, “VL-ROUTE: A Cross-Layer Routing
aware applications,” IEEE Communications Surveys Tutorials, vol. 18, Protocol for Visible Light Ad Hoc Network,” in Proc. of IEEE Symp.
no. 1, pp. 525–552, Firstquarter 2016. on a World of Wireless, Mobile, and Multimedia Networks (WoWMoM),
[13] Xiaojun Lin, N. B. Shroff, and R. Srikant, “A tutorial on cross-layer Washington D.C., USA, June 2019.
optimization in wireless networks,” IEEE Journal on Selected Areas in [36] N. T. Dang and V. V. Mai, “A phy/mac cross-layer analysis for ieee
Comms, vol. 24, no. 8, pp. 1452–1463, 2006. 802.15. 7 uplink visible local area network,” IEEE Photonics Journal,
[14] S. Shakkottai, T. Rappaport, and P. Karlsson, “Cross-layer design for vol. 11, no. 3, pp. 1–17, 2019.
wireless networks,” IEEE Communications Magazine, vol. 41, no. 10, [37] J. Jagannath and T. Melodia, “An Opportunistic Medium Access Control
pp. 74–80, 2003. Protocol for Visible Light Ad Hoc Networks,” in Proc. of International
[15] J. Jagannath, A. Jagannath, Z. Henney, J. Kane, N. Biswas, T. Gwin, Conference on Computing, Networking and Communications (ICNC),
and A. Drozd, “Fieldable cross-layer optimized embedded software Maui, Hawaii, USA, March 2017.
defined radio is finally here!” in Proc. of IEEE Military Communications [38] L. Ding, T. Melodia, S. Batalama, J. Matyjas, and M. Medley, “Cross-
Conference (MILCOM), Dec 2021. layer Routing and Dynamic Spectrum Allocation in Cognitive Radio Ad
[16] J. Jagannath, “Cross-layer optimized networking for next-generation 5g Hoc Networks,” IEEE Transactions on Vehicular Technology, vol. 59,
ad hoc networks,” Ph.D. dissertation, Northeastern University, 2019. pp. 1969–1979, May 2010.
[17] L. D. Mendes and J. J. Rodrigues, “A survey on cross-layer solutions [39] L. Ding, K. Gao, T. Melodia, S. Batalama, D. Pados, and J. Maty-
for wireless sensor networks,” Journal of Network and Computer Appli- jas, “All-spectrum Cognitive Networking through Jointly Optimal Dis-
cations, vol. 34, no. 2, pp. 523 – 534, 2011. tributed Channelization and Routing,” IEEE Transactions on Wireless
[18] X. Wang and K. Kar, “Cross-layer rate optimization for proportional Communications, vol. 12, no. 11, pp. 5394–5405, November 2013.
fairness in multihop wireless networks with random access,” IEEE [40] C. She, C. Yang, and T. Q. S. Quek, “Cross-layer optimization for ultra-
Journal on Selected Areas in Comms, vol. 24, pp. 1548–1559, 2006. reliable and low-latency radio access networks,” IEEE Transactions on
[19] I. Nosheen, S. A. Khan, and F. Khalique, “A mathematical model for Wireless Communications, vol. 17, no. 1, pp. 127–141, 2018.
cross layer protocol optimizing performance of software-defined radios [41] J. Jagannath, S. Furman, T. Melodia, and A. Drozd, “Design and
in tactical networks,” IEEE Access, vol. 7, pp. 20 520–20 530, 2019. experimental evaluation of a cross-layer deadline-based joint routing
[20] J. Jagannath, T. Melodia, and A. Drozd, “DRS: Distributed Deadline- and spectrum allocation algorithm,” IEEE Transactions on Mobile
Based Joint Routing and Spectrum Allocation for Tactical Ad-hoc Net- Computing, vol. 18, no. 8, pp. 1774–1788, 2019.
works,” in Proc. of IEEE Global Communications Conference (GLOBE- [42] E.-C. Park, D.-Y. Kim, H. Kim, and C.-H. Choi, “A cross-layer approach
COM), Washington, DC USA, December 2016. for per-station fairness in tcp over wlans,” IEEE Transactions on Mobile
[21] T. Zhu, X. Chen, L. Chen, W. Wang, and G. Wei, “Gclr: Gnn-based Computing, vol. 7, no. 7, pp. 898–911, 2008.
cross layer optimization for multipath tcp by routing,” IEEE Access, [43] C. Xu, Z. Li, J. Li, H. Zhang, and G.-M. Muntean, “Cross-layer
vol. 8, pp. 17 060–17 070, 2020. fairness-driven concurrent multipath video delivery over heterogeneous
[22] M. J. Herrmann and G. G. Messier, “Cross-layer lifetime optimization wireless networks,” IEEE Transactions on Circuits and Systems for
for practical industrial wireless networks: A petroleum refinery case Video Technology, vol. 25, no. 7, pp. 1175–1189, 2015.
study,” IEEE Transactions on Industrial Informatics, vol. 14, no. 8, pp. [44] C. Dai, W. Li, L. Wang, M. Song, Y. Zhang, and J. Xiao, “A cross layer
3559–3566, 2018. routing protocol for green communication network,” in IET International
[23] J. Jagannath and S. Furman and A. Jagannath and A. Drozd, “Energy Conference on Communication Technology and Application (ICCTA
Efficient Ad Hoc Networking Devices for Off-the-Grid Public Safety 2011), 2011, pp. 910–914.
Networks,” in Proc. of IEEE Consumer Communications & Networking [45] H. B. Salameh, M. B. Irshaid, A. A. Ajlouni, and M. Aloqaily, “Energy-
Conference (CCNC), Las Vegas, NV, USA, January 2019. efficient cross-layer spectrum sharing in cr green iot networks,” IEEE
[24] S. Barmpounakis, N. Maroulis, M. Papadakis, G. Tsiatsios, D. Soukaras, Transactions on Green Communications and Networking, pp. 1–1, 2021.
and N. Alonistioti, “Network slicing - enabled ran management for 5g: [46] M. Al-Jemeli and F. A. Hussin, “An energy efficient cross-layer net-
Cross layer control based on sdn and sdr,” Computer Networks, vol. 166, work operation model for ieee 802.15.4-based mobile wireless sensor
p. 106987, 2020. networks,” IEEE Sensors Journal, vol. 15, no. 2, pp. 684–692, 2015.
[25] L. Zhang and X. Zhou, “Joint cross-layer optimised routing and dynamic [47] D. Xue and E. Ekici, “Delay-Guaranteed Cross-Layer Scheduling in
power allocation in deep space information networks under predictable Multihop Wireless Networks,” IEEE/ACM Transactions on Networking,
contacts,” IET Communications, vol. 7, no. 5, pp. 417–429, 2013. vol. 21, no. 6, pp. 1696–1707, Dec 2013.
[26] A. Jagannath, J. Jagannath, and A. Drozd, “Artificial Intelligence- [48] X. Liu, K. Hua, Z. Chen, A. S. Alghamdi, and M. N. Ali, “An efficient
based Cognitive Cross-layer Decision Engine for Next-Generation Space cross-layer approach for throughput-maximal and delay-minimal green
 20

vehicular networks,” in 2018 International Conference on Computing, [69] S. Liu, E. Ekici, and L. Ying, “Scheduling in multihop wireless networks
Networking and Communications (ICNC), 2018, pp. 652–658. without back-pressure,” IEEE/ACM Transactions on Networking, vol. 22,
[49] I. Jemili, D. Ghrab, A. Belghith, and M. Mosbah, “Cross-layer no. 5, pp. 1477–1488, Oct 2014.
adaptive multipath routing for multimedia wireless sensor networks [70] K. Hao, Z. Jin, H. Shen, and Y. Wang, “An efficient and reliable geo-
under duty cycle mode,” Ad Hoc Networks, vol. 109, p. 102292, 2020. graphic routing protocol based on partial network coding for underwater
[Online]. Available: https://www.sciencedirect.com/science/article/pii/ sensor networks,” Sensors, vol. 15, no. 6, pp. 12 720–12 735, 2015.
S157087052030651X [71] J. Jagannath, A. L. Drozd, A. Jagannath, S. R. Furman, and A. K. Burger,
[50] M. Cheng, Q. Ye, and L. Cai, “Cross-layer schemes for reducing “Routing data through distributed communications network,” Jul. 21
delay in multihop wireless networks,” IEEE Transactions on Wireless 2020, US Patent 10,721,162.
Communications, vol. 12, no. 2, pp. 928–937, 2013.
[51] H. Xiong, R. Li, A. Eryilmaz, and E. Ekici, “Delay-Aware Cross-Layer
Design for Network Utility Maximization in Multi-Hop Networks,”
IEEE Journal on Selected Areas in Communications, vol. 29, no. 5,
pp. 951–959, May 2011.
[52] I. Jemili, D. Ghrab, A. Belghith, M. Mosbah, and S. Al- Jithin Jagannath (SM’19) is the Chief Technology
Ahmadi, “Cross-layer multipath approach for critical traffic in Scientist and Founding Director of the Marconi-
duty-cycled wireless sensor networks,” Journal of Network and Rosenblatt AI/ML Innovation Lab at ANDRO Com-
Computer Applications, vol. 191, p. 103154, 2021. [Online]. Available: putational Solutions. He is also the Adjunct As-
https://www.sciencedirect.com/science/article/pii/S1084804521001685 sistant Professor in the Department of Electrical
[53] S. R. Lahane and K. N. Jariwala, “Secured cross-layer cross-domain Engineering at the University at Buffalo, State Uni-
routing in dense wireless sensor network: A new hybrid based clustering versity of New York. Dr. Jagannath received his B.
approach,” International Journal of Intelligent Systems, vol. 36, pp. 3789 Tech in Electronics and Communication from Kerala
– 3812, 2021. University; M.S. degree in Electrical Engineering
[54] American National Standard ANSI/ISA-100.11a-2011, “ANSI/ISA- from University at Buffalo, The State University
100.11a-2011 Wireless Systems for Industrial Automation: Process of New York; and received his Ph.D. degree in
Control and Related Applications,” May 2011. Electrical Engineering from Northeastern University. He is an IEEE Senior
[55] D. Chen, M. Nixon, and A. Mok, WirelessHART: Real-Time Mesh member and serves as IEEE Industry DSP Technology Standing Committee
Network for Industrial Automation, 1st ed. Springer Publishing member. He also serves on the Federal Communication Commission’s (FCC)
Company, Incorporated, 2010. Communications Security, Reliability, and Interoperability Council (CSRIC
[56] P. Shome, M. Yan, S. M. Najafabad, N. Mastronarde, and A. Sprintson, VIII) Working Group 1. Dr. Jagannath was the recipient of the 2021 IEEE
“Crossflow: A cross-layer architecture for sdr using sdn principles,” in Region 1 Technological Innovation Award with the citation, "For innovative
2015 IEEE Conference on Network Function Virtualization and Software contributions in machine learning techniques for the wireless domain”.
Defined Network (NFV-SDN), 2015, pp. 37–39. Dr. Jagannath heads several of the ANDRO’s research and development
[57] E. Demirors, G. Sklivanitis, T. Melodia, and S. N. Batalama, “RcUBe: projects in the field of Beyond 5G, signal processing, RF signal intelligence,
Real-Time Reconfigurable Radio Framework with Self-Optimization cognitive radio, cross-layer ad-hoc networks, Internet-of-Things, AI-enabled
Capabilitites,” in Proc. of IEEE Intl. Conf. on Sensing, Communication, wireless, and machine learning. He has been the lead and Principal Investigator
and Networking (SECON), Seattle, WA, June 2015. (PI) of several multi-million dollar research projects. This includes a Rapid
Innovation Fund (RIF) and several Small Business Innovation Research
[58] J. Jagannath, H. Saarinen, W. Timothy, J. O’Brien, S. Furman, T. Melo-
(SBIR)s for several customers including the U.S. Army, U.S Navy, Department
dia, and A. Drozd, “COmBAT: Cross-layer Based Testbed with Analysis
of Homeland Security (DHS), United States Special Operations Command
Tool Implemented Using Software Defined Radios,” in Proc. of IEEE
(SOCOM). He is currently leading several teams developing commercial
Conf. on Military Comm. (MILCOM), Baltimore, MD, November 2016.
products such as SPEARLink™, DEEPSpec™ among others. He is the
[59] S. Soltani, Y. E. Sagduyu, S. Scanlon, Y. Shi, J. H. Li, J. Feldman, and inventor of 11 U.S. Patents (granted, pending, and provisional). He has been
J. D. Matyjas, “Cross-layer design and sdr implementation of dsa, back- invited to give various talks including Keynote on the topic of machine
pressure routing and network coding,” arXiv preprint arXiv:1912.05658, learning and Beyond 5G wireless communication. He has been invited to
2019. serve on the Technical Program Committee for several leading technical
[60] G. Sklivanitis, E. Demirors, A. Gannon, S. N. Batalama, D. A. Pados, conferences.
and S. N. Batalama, “All-spectrum cognitive channelization around
narrowband and wideband primary stations,” in Proc. of IEEE Global
Communications Conf. (GLOBECOM), San Diego, CA, December 2015.
[61] G. Kakkavas, K. Tsitseklis, V. Karyotis, and S. Papavassiliou, “A
software defined radio cross-layer resource allocation approach for
cognitive radio networks: From theory to practice,” IEEE Transactions
on Cognitive Communications and Networking, vol. 6, no. 2, pp. 740– Anu Jagannath received her MS degree (2013)
755, 2020. from State University of New York at Buffalo in
[62] “ORCA-Orchestration and Reconfiguration Control Architecture,” Electrical Engineering. She is currently a Senior
Accessed- Aug 2021. [Online]. Available: https://www.orca-project.eu/ Scientist at ANDRO Computational Solutions, LLC
[63] “Iris-The Reconfigurable Radio Testbed,” Accessed- Aug 2021. and serves as the Associate Director of Marconi-
[Online]. Available: http://iristestbed.eu/ Rosenblatt AI/ML Innovation Lab at ANDRO. Her
[64] “Open-Access Research Testbed for Next-Generation Wireless Networks research focuses on MIMO communications, Deep
(ORBIT),” Accessed- Aug 2021. [Online]. Available: https://www. Machine Learning, Reinforcement Learning, Adap-
orbit-lab.org/ tive signal processing, Software Defined Radios,
[65] J. Jagannath and S. Furman and A. Jagannath and L. Ling and A. spread spectrum systems, LPI/LPD communications,
Burger and A. Drozd, “HELPER: Heterogeneous Efficient Low Power spectrum sensing, adaptive Physical layer, and cross-
Radio for Enabling Ad Hoc Emergency Public Safety Networks,” Ad layer techniques, medium access control and routing protocols, underwater
Hoc Networks (Elsevier), vol. 89, pp. 218 – 235, 2019. wireless sensor networks, signal intelligence and so on. She serves as author
[66] Sidekiq-Z2. Epiq Solutions. Accessed July 19, and coauthor for book chapters and multiple research publications in journals
2021. [Online]. Available: https://epiqsolutions.com/static/ and conference proceedings. She has rendered her reviewing service for
Epiq-Solutions-Sidekiq-Z2-8a9337a3ad0d6387032e52c2db623625.pdf conferences such as IEEE Annual Consumer Communications & Networking
[67] X. Chen, Y. Xu, and A. Liu, “Cross layer design for optimizing Conference (CCNC) and IEEE International Workshop on Signal Processing
transmission reliability, energy efficiency, and lifetime in body sensor Advances in Wireless Communications (SPAWC). She is the co-Principal
networks,” Sensors, vol. 17, no. 4, p. 900, 2017. Investigator (co-PI) and Technical Lead in multiple Rapid Innovation Fund
[68] L. Tassiulas and A. Ephremides, “Stability properties of constrained (RIF) and SBIR/STTR efforts involving developing embedded MIMO solu-
queueing systems and scheduling policies for maximum throughput in tions, deep and reinforcement learning for wireless communications, signal
multihop radio networks,” in Proc. of IEEE Conference on Decision and intelligence, and mesh networking.
Control (CDC), Honolulu, HI, December 1990.
 21

Justin Henney received his BS from the Rochester Andrew Drozd received his Ph.D. from St. John
Institute of Technology in Electrical Engineering in Fisher College in Executive Leadership in Science &
2019. He is currently an Associate Scientist/Engi- Technology Towards Advancements in Human and
neer at ANDRO Computational Solutions, and has Machine Learning Using Theoretic Frameworks for
worked on developing the SPEARLink™ and AR- AI/ML Applications. Dr. Drozd has extensive expe-
ROWLink™ products. rience in electromagnetics science and technology
with expertise in emerging spectrum technologies,
radar cross section (RCS), computational electro-
magnetics (CEM), radar and multisensor data fusion,
multitarget identification/tracking, target discrimina-
tion, cyber and electronic warfare (EW), and cosite
modeling and simulation through his various working relationships over the
past 43 years with DoD and Federal government agencies, including the U.S.
Army, U.S. Air Force, U.S. Navy, Office of the Secretary of Defense (OSD),
Missile Defense Agency (MDA), Defense Information Systems Agency /
Department of Spectrum Operations (DISA/DSO), and Department of Home-
land Security (DHS). Over the past several years, Dr. Drozd has led efforts
to develop and mature new technologies under Small Business Innovative
Research / Small-business Technology Transfer Research (SBIR/STTR) and
Commercialization Readiness Program (CRP) contracts.
Tyler Gwin received their B.S. degree in Computer Dr. Drozd is an iNARTE (international Association of Radio and Telecom-
and Information Science from SUNY Polytechnic munications Engineers) Certified Engineer (1988-present); IEEE Life Fellow;
Institute. He is currently an Associate Scientist/Engi- past president of the IEEE EMC Society (2006-2007); past chairman of the
neer at ANDRO Computational Solutions, LLC. IEEE EMC-S Standards Development and Education Committee (2007-2009);
During his work at ANDRO, he has been part of member of the FCC’s Communications, Security, Reliability, Interoperability
the team developing SPEARLink™ software, along Council (CSRIC VIII - 2021-2023); member of the Alpha Gamma Omicron
with the research and development of small Un- Chapter of Kappa Delta Pi Honor Society in Education, board member of the
manned Aerial Systems. Tyler has also assisted in Project Fibonacci® Foundation, Inc. focused on STEM leadership education
the authoring of a book chapter titled "Deep Learn- for workforce development; and a 2019 inductee in the Rome Academy of
ing and Reinforcement Learning for Autonomous Sciences Hall of Fame. He holds 10 patents and patents pending including
Unmanned Aerial Systems: Roadmap for Theory to the areas of spectrum governance, signal detection/classification, and block
Deployment." chain auditing.

Zackary Kane received a B.S. in Electrical and

Computer Engineer from the State University of
New York Polytechnic Institute. Zack is currently
a Radio Frequency Engineer/Associate Scientist I
at ANDRO Computational Solutions in ANDRO’s
Marconi-Rosenblatt lab. He has worked on two
Small Business Innovation Research (SBIR) projects
and the STRAIN Rapid Innovation Fund (RIF)
project. While on STRAIN, Zack assisted in range
testing the SPEARLink™ radio platform and worked
on documentation for the SPEARLink™ radio. He
is currently working on the DeepeRFind Phase I SBIR, generating an database
of commercial radio devices, and the I-ROAM RIF, testing and improving the
IROAM CDMA physical layer.

Noor Biswas received his BS degree in Computer

Engineering from the State University of New York
Polytechnic Institute in 2020. He is currently an
Associate Scientist/Engineer at ANDRO Computa-
tional Solutions, and has contributed to projects
involving software-defined radio networks.