HR001117S0043.pdf

HR001117S0043

Microsystems Technology Office Broad Agency Announcement

Radio Frequency Machine Learning Systems (RFMLS)

HR001117S0043

August 11, 2017

Table of Contents

PART I: OVERVIEW INFORMATION

PART II: FULL TEXT OF ANNOUNCEMENT

I. Funding Opportunity Description A. Background B. Program Description

1. TA1 and TA2: Algorithms and Architectures

2. TA3: RF System Integrator and Demonstrator

II. Award Information A. General Award Information B. Fundamental Research

III. Eligibility Information A. Eligible Applicants

1. Federally Funded Research and Development Centers (FFRDCs) and Government Entities

2. Non-U.S. Organizations and/or Individuals B. Organizational Conflicts of Interest C. Cost Sharing/Matching D. Associate Contractor Agreement Clause E. Other Eligibility Criteria

IV. Application and Submission Information A. Address to Request Application Package B. Content and Form of Application Submission

1. Full Proposal Format

2. Proprietary Information

3. Security Information

a. Unclassified Submissions

b. Classified Submissions

4. Disclosure of Information and Compliance with Safeguarding Covered

Defense Information Controls

5. Human Research Subjects/Animal Use

6. Approved Cost Accounting System Documentation

7. Section 508 of the Rehabilitation Act (29 U.S.C. § 749d)/FAR 39.2

8. Small Business Subcontracting Plan

9. Intellectual Property

a. For Procurement Contracts

b. For All Non-Procurement Contracts

10. Patents

11. System for Award Management (SAM) and Universal Identifier Requirements

12. Funding Restrictions

C. Submission Information

1. Submission Dates and Times

a. Full Proposal Date

b. Frequently Asked Questions (FAQ)

2. Proposal Submission Information

a. For Proposers Requesting Contracts or Other Transaction Agreements

b. Classified Submission Information

3. Other Submission Requirements

V. Application Review Information A. Evaluation Criteria

1. Overall Scientific and Technical Merit

2. Potential Contribution and Relevance to the DARPA Mission

3. Proposer’s Capabilities and/or Related Experience

4. Cost Realism

5. Plans and Capability to Accomplish Technology Transition

B. Review and Selection Process

1. Review Process

2. Handling of Source Selection Information

3. Federal Awardee Performance and Integrity Information (FAPIIS)

VI. Award Administration Information A. Selection Notices B. Administrative and National Policy Requirements

1. Meeting and Travel Requirements

2. FAR and DFARS Clauses

3. Controlled Unclassified Information (CUI) on Non-DoD Information Systems

4. Representations and Certifications

C. Reporting D. Electronic Systems

1. Wide Area Work Flow (WAWF)

2. i-Edison

VII. Agency Contacts VIII. Other Information

A. Proposers Day B. Protesting

ATTACHMENT 1: Cost Volume Proposer Checklist ATTACHMENT 2: Proposal Summary Chart Template

PART I: OVERVIEW INFORMATION

Federal Agency Name: Defense Advanced Research Projects Agency (DARPA), Microsystems Technology Office (MTO)

Funding Opportunity Title: Radio Frequency Machine Learning Systems (RFMLS) Announcement Type: Initial Announcement Funding Opportunity Number: HR001117S0043 Catalog of Federal Domestic Assistance Numbers (CFDA): 12.910 Research and

Technology Development Dates: (All times listed herein are Eastern Time) o Posting Date: August 11, 2017 o Proposers Day: August 31, 2017 (Registration: http://www.cvent.com/d/55qt96) o FAQ Submission Deadline: September 19, 2017 o Proposal Due Date: October 10, 2017 o Estimated period of performance start: April 2018

Concise description of the funding opportunity: The goal of the RF Machine Learning Systems (RFMLS) program is to develop the foundations for applying modern data-driven Machine Learning to the RF Spectrum domain as well as to develop practical applications in emerging spectrum problems, which demand vastly improved discrimination performance over today's hand-engineered RF systems. Ultimately, these innovations will result in a new generation of RF systems that are goal-driven and can learn from data rather than being hand-engineered by experts.

Anticipated Funding Available for Award: Multiple Technical Area 1 and 2 awards are expected with an approximate average award amount of $1.0-4.0M per year per award. Multiple Technical Area 3 awards may be made with an approximate average award amount of $1.0-1.5M per year per award.

Anticipated individual awards: Multiple awards are anticipated in Technical Areas 1 and 2. One or more awards are anticipated in Technical Area 3.

Anticipated funding type: 6.2 Types of instruments that may be awarded: Procurement contract or other transaction.

Agency contact:

o Paul Tilghman, Program Manager HR001117S0043@darpa.mil

DARPA/MTO

ATTN: HR001117S0043

675 North Randolph Street Arlington, VA 22203-2114 http://www.cvent.com/d/55qt96 mailto:name@darpa.mil

PART II: FULL TEXT OF ANNOUNCEMENT

I. Funding Opportunity Description

The Defense Advanced Research Projects Agency (DARPA) often selects its research efforts through the Broad Agency Announcement (BAA) process. This BAA is being issued, and any resultant selection will be made, using the procedures under Federal Acquisition Regulation (FAR) 6.102(d)(2) and 35.016 and 2 C.F.R. § 200.203. Any negotiations and/or awards will use procedures under FAR 15.4, Contract Pricing. Proposals received as a result of this BAA shall be evaluated in accordance with evaluation criteria specified herein through a scientific review process.

DARPA BAAs are posted on the Federal Business Opportunities (FedBizOpps) website, http://www.fbo.gov/. The following information is for those wishing to respond to the BAA.

The Microsystems Technology Office at DARPA seeks innovative proposals exploring the intersection between the RF spectrum domain and modern data-driven machine learning (such as Deep Learning). Such efforts may include tailoring existing machine learning approaches to handle the unique aspects of the RF spectrum modality or exploration into new architectures, learning algorithms, etc. better suited to this modality. Proposed research should investigate innovative approaches that enable revolutionary advances in science, devices, or systems.

Specifically excluded is research that primarily results in evolutionary improvements to the existing state of practice.

A. Background

It has been nearly 20 years since the first emergence of the term "Cognitive Radio," which introduced the field of wireless radio frequency (RF) practitioners to the domain of Artificial Intelligence (AI). Since that time, the field of AI has experienced a renaissance fueled by modern data-driven Machine Learning (ML), predominantly under the umbrella moniker of Deep Learning. In contrast to the expert-engineered processing and decision making of the previous generation of AI, ML-based techniques are able to learn to perform a task from raw sensory input. ML techniques have advanced the fields of image and speech recognition from toy problems to usage in everyday life. In contrast, the use of AI in the RF domain however has not kept pace with these advancements in ML, and little work has been conducted to explore the intersection of traditional RF signal processing and machine learning.

The goal of the Radio Frequency Machine Learning Systems (RFMLS) program is to develop the foundations for applying modern ML to the RF Spectrum domain and develop practical applications in emerging spectrum problems, which demand vastly improved discrimination performance over today's hand-engineered first generation Cognitive RF systems. Ultimately, these innovations will result in a new generation of RF systems that are goal-driven and can learn from data, allowing experts to focus on the overall system capability and not individual sub-system performance.

http://www.fbo.gov/

Traditionally, ML research has focused on problems in the domains of vision, hearing, and language, as these are human abilities displaying intelligent behavior. Progress in these areas has been significant in recent years, surpassing human abilities in some ways. Though the RF environment presents familiar ground for ML researchers, there are some important differences that make RFML unique. Compared to other domains, the RFML data rate is much higher. A high definition video stream at 60 Hz could be on the order of 3 Gbps (ignoring coding compression savings). 1 GHz of RF spectrum data has approximately 10 times that data rate, and is not nearly as amenable to decimation in order to improve processing. Additionally, RF waveforms are typically captured and represented as complex numbers, underscoring the importance of both amplitude and phase of the signal. Although there has been interest recently in complex-valued neural networks, the technology for learning naturally in the complex plane is not fully developed and relies on treating complex variables as two real numbers. The RF environment offers a more compelling reason for developing a native complex-valued approach.

Other prominent differences between current ML applications and the RF domain exist and must be addressed by the RFMLS Program.

Many of the best-performing modern ML algorithms began as some variation of the biology attempting to be mimicked (e.g., vision, hearing, etc.) combined with best practices discovered over time through trial and error. While no analogous biology for RF exists, the structure of RFMLS is nonetheless inspired by the capabilities of biological systems. It is expected that the RFMLS program will create new or adapt existing, ML practices, structures, and algorithms specialized for the RF domain. Four key capabilities form the foundation for an eventual RFML System as well the overall structure of the RFMLS program.

1. Feature Learning Applications of first-generation AI to the spectrum currently depend on hand-engineered features which an expert has selected based on the belief that they best describe RF signals pertinent to a specific RF task. Conversely, Deep Learning applied in other domains has achieved excellent performance in vision and speech applications by learning features similar to those learned by the brain from sensory data. Recently it has been shown that machine learning of RF features has the potential to transform spectrum problems in a way similar to other domains. Specifically, it is expected that RFML Systems will be capable of learning the appropriate features to describe RF signals and associated properties from training data.

2. Attention & Saliency Next-generation RF systems must evolve from characterizing MHz of spectrum to characterizing GHz of spectrum. To effectively make this step, RFML Systems must have a sense of which signals are important, focusing processing resources on those signals, and conserve resources by disregarding other signals that are unimportant to a particular task. Humans are exquisite at consuming, prioritizing, and processing visual and auditory information. A major contributor to this ability is a sophisticated sense of importance defined by attention and saliency. Top-down attention is a goal-driven mechanism, which causes us to focus our cognitive processing on visual information most pertinent to a task at hand. For example, while looking for faces, humans will tend to focus their attention at eye-level since the floor is unlikely to contain many faces. Adhering strictly to goal-driven attention, however, would ultimately be myopic, missing new information relevant to the task at hand. Hence attention is complemented by a bottom-up

(e.g. data-driven) mechanism called saliency, which enables us to capture changes or contrast differences in the visual or auditory scene.

Top-down attention in vision and speech is important to select the relevant low-level features needed for a classification or decision while removing distractors. Bottom-up saliency is important to call for top-down and cognitive attention due to, for example, a bright moving object or a loud clap. Similarly, for understanding an RF scene, stored RF concepts such as signals and transmitter types can be used to identify RF objects and model behavior. ML methods that include recurrent structures have been shown to create attentional functions in other modalities creating a path for such functions in an RFML System.

3. Autonomous RF Sensor Configuration Unlike traditional RF systems, visual processing is an autonomous sensory-motor task. If the environment is bright our eyes adjust. Saccadic vision directs eye movement to capture the most relevant portions of the visual scene. Much of the early research in Adaptive RF Systems focused on the analog RF electronics (i.e., the RF front end), building systems that can be reconfigured, so that the operating characteristics are not set in stone at design time. Similarly recent advances have resulted in large-scale, all-digital RF systems that are intrinsically flexible.

These advances are largely untapped because there has been little focus on coupling this adaptability with the necessary intelligence to reconfigure the RF sensor for better performance of the overall system task.

Learning for autonomous control has revolutionized game playing and self-driving cars.

Although the sensors and actuators are different in RF applications, an RFML system can learn to optimize the huge number of possible configurations (analog electronics, beam steering angle, bandwidth sensitivity, position, etc.). ML methods that combine sensory processing (e.g.

Convolutional Neural Networks) and decision-making when also combined with appropriate training methods (e.g. Reinforcement Learning) have enabled machine-learning systems to take on tasks like the game of Go, previously thought to be the domain of humans for at least the next decade. The space of RF configurations is combinatorically large but, as in the game of Go, susceptible to ML. RFML Systems will be capable of autonomously reconfiguring the sensor over a large control space to improve overall task performance.

4. Waveform Synthesis Tasks such as waveform generation in present-day RF systems amount to decisions on a small set of discrete choices – use waveform A, or use waveform B. This fails to achieve true waveform synthesis – the ability to generate wholly new waveforms. Discrete selection approaches are akin to a speech-generation system preprogrammed with a discrete set of words, which it can choose between. Such a system could not pronounce new words or apply more nuanced concepts such as syllabic stress common in modern speech synthesis systems. Similar challenges exist in the RF domain in coping with very large decision spaces. ML approaches, which enable unsupervised or semi/self-supervised learning of efficient waveform encodings, will enable RFML Systems to learn fundamentally new waveforms able to achieve one or more simultaneous objectives.

B. Program Description

The goal of the RFMLS Program is to develop and demonstrate four constituent capabilities for RF systems:

1. Feature Learning – the ability to learn features necessary to perform a discrimination task directly from sensor data

2. Attention and Saliency – the ability to have top-down (task-driven) attention guide which RF samples are ultimately most important to the system’s task, complemented by a bottom-up (data-driven) mechanism able to direct attention to new signals of interest

3. Autonomous RF Sensor Configuration – the ability to autonomously reconfigure an RF sensor (e.g., center frequency tuning, beamforming direction, analog sensor reconfiguration) to create the RF data required to improve overall task performance

4. Waveform Synthesis – the ability of an RF system to learn to transmit entirely new waveforms to achieve a task

The RFMLS Program has three technical areas (TAs). TA1 and TA2 will develop RF machine-learning algorithms and architectures. TA1 and TA2 each have two tasks emphasizing one of the four constituent RFMLS capabilities. While each of the four capability areas is emphasized in one task, the capability areas are not orthogonal. For example, RF feature learning is likely to be required in all tasks, but the RF Feature Learning task (Task 1A) is considered to stress this capability more than the other tasks. TA3 will provide an RF system for integration and demonstration. The following TA descriptions detail the interaction required between performers to support development, integration, testing, and demonstration.

The RFMLS Program consists of three Phases each lasting 12 months. TA1 does not extend beyond Phase 2. Program Phases 2 and 3 should be proposed as a separate option with their own SOW (statement of work) and cost. TA1 and TA2 proposers should provide separate options for each of the two capability tasks. Proposers bidding to both TA1 and TA2 should describe any cost-savings yielded if two awards are made.

Figure 1 – Anticipated program schedule

To avoid organizational conflict of interest and to ensure objectivity in the System Integrator role, a proposer awarded under TA3 cannot be selected for any portion of TA1 of TA2, either as a prime, subcontractor, or in any other capacity from an organizational to individual level. For the purposes of this BAA, DARPA will consider proposals from the same organization with different CAGE codes to be separate organizations. While proposers may submit proposals for both TA3 and TA1/TA2, the decision as to which proposal(s) to consider for award is at the discretion of the Government.

Proposers should ensure that the topics in Section IV.4.2.C (Technical Approach) are addressed in their proposals.

Separate proposals must be submitted for each TA proposed.

1. TA1 and TA2: Algorithms and Architectures

TA1 is motivated by the need for enhanced wireless security in the IoT (internet of things), while TA2 is motivated by spectrum monitoring to support spectrum sharing. A proposer may propose under either or both TA1 and TA2, and each proposal will be evaluated independently. TA1 and TA2 each contain two tasks emphasizing two of the four desired RFMLS capabilities. Proposers must address both tasks of the TA.

TA1 and TA2 proposed solutions are expected to have two modes of operation: training and evaluation. In training mode the system will consume the datasets provided in order to learn its tasks. Training mode is strictly an offline process. In evaluation mode, online incremental learning is not strictly a requirement, however unsupervised clustering is an online requirement of some tasks.

TA1 and TA2 proposers are responsible for the computational platform both for training and evaluation. Due to the foundational nature of the RFMLS program, it is not necessary that the computational platform meet any specific form factor. Real-time evaluation is a requirement as detailed below.

a) TA1 RF Forensics

Background The IoT relies on each networked device to honestly present a digital ID (identification). As 2 illustrates, digital IDs (in software and firmware) can be cloned, so IoT security would benefit if it were supplemented by a unique physical ID intrinsic to the device. RF transmitters are naturally imperfect devices due to the tolerances in manufacturing of the analog electronics. A “radiofrequency fingerprint” (RFFP) for each transmitting device can be measured. The communications security literature has many examples of the RFFP concept. The typical research centers around a few nominally identical transmitters of a single protocol type and studies their transmit signals in order to hand-engineer features that discriminate the individual transmitters. While successful as proof of concept demonstrations, hand-engineered-feature discriminators have not been shown to support the number of devices expected in IoT populations. Similarly, hand-engineering approaches which focus on only a single type of RF system (e.g. WiFi) will not handle the diversity of wireless protocols emerging in the IoT.

Figure 2 – TA1 supports “Bob” through RF fingerprinting

A detailed schedule must be provided by TA1 proposers which details the interaction between the development associated with Tasks 1A and 1B.

Task 1A: RF Feature Learning

Task Description TA1 performers must develop a trainable system (algorithms, not hardware) that can distinguish between 10,000+ RF devices that are nominally identical. Since these signals are indistinguishable using conventional hand-engineered RF signal features (e.g. modulation type, rate), the Task 1A system must learn more abstract features which are imparted to the intended RF signal by the RF transmit chain and associated hardware. The trainable system must be general such that it is applicable to multiple wireless communications protocols. The algorithms should be robust to training conditions, which include varied signal-to-noise level, adjacent channel interference, co-channel interference, and wireless channel effects imparted upon the signals of interest.

Proposers should describe their algorithmic approach to learning RFFP features. In doing so proposers should describe how their system learns features and what type of features of the transmitter are learned. The generalizability of the approach, how it enables fingerprints to be learned for different waveform and protocol types, should be clearly identified. Additionally the differences between training and evaluation operation of the system should also be described.

Task 1A is comprised of two 12 month phases. In Phase 1 the GFI (Government Furnished Information) dataset will contain at least 2 protocols known to the performers. Evaluation conducted by the government team will include evaluation against a previously undisclosed communications waveform. During Phase 1 there is no required time limit for processing during online evaluation.

In Phase 2 the dataset will grow to include data gathered using different receive hardware as well as potential additional wireless protocols. At the end of Phase 2 the performer must demonstrate “real time” evaluation. During this final demonstration the signal samples will be streamed to the Task 1A system at the RF sampling rate (e.g. 200 MHz). Evaluation of a 1s signal sample should be completed within 2s of the start of that sample. At the end of Phase 2, Task 1A performers will demonstrate that their algorithms work when applied to real-time streaming data from TA3 hardware.

Government Furnished Information (GFI) At the start of Phase 1, DARPA will provide GFI digitized RF signal samples for training and testing. Since the signal samples are digital, TA1 performers do not require (but may benefit from) prior RF experience. The signal database will include multiple signal types (e.g. WiFi, Bluetooth), multiple examples from each device, and as many devices as possible. Proposers should expect that a mix of device manufacturers are present in the data. A typical dataset will have signals sampled for 0.1s at 200 MHz with each example as a complex number having 12 bits IQ (in-phase/real component, quadrature-phase/imaginary component) data. This is 60 MB per example, or 6 TB for a 100,000 example dataset. To the degree possible, signals in the database will be “open-air,” that is the transmitted signal propagates through a real wireless environment. Proposers should not plan to create or use their own signal samples.

A training dataset will contain one or more examples from each transmitter in the database, along with a unique identity label (truth data) for the transmitter. After training, the evaluation dataset will contain new examples from many of the transmitters in the training database, plus signals from new transmitters. The Task 1A system must either match the signal to a transmitter in the training database and report an old ID number, or declare the signal to be transmitted by a new device and assign a new ID number.

During Phase 2 an additional dataset will be provided which includes signals captured by TA3 provided hardware. This dataset may also increase the total number of communications types.

Performance Metrics Correct identification of the transmitter is counted as a “detection”, as is correct identification of the fact that the transmitter is new. The following table shows performance goals in terms of PD, the probability of detection.

Description Threshold Goal Task 1A

PD is the probability of correctly identifying a transmitter during a future encounter

PD > 0.90 with a population of 1k transmitters

PD > 0.95 with a population of 10k transmitters

Task 1B: RF Waveform Synthesis

Task Description Task 1A relies on learning the signal imperfections naturally present and unintentionally imparted when transmitters send a pre-existing standards-based communications signal. Under Task 1B, performers will develop a transmitting RF system that learns to intentionally exploit the intrinsic signaling characteristics of the transmitter in order to enhance the performance of a Task 1A system’s ability to properly ID the transmitter.

Performers will be learning to custom tailor a waveform with two objectives:

1. Communication – The waveform learned still must successfully convey the intended information at a nominal rate

2. Identifiability – The waveform learned is uniquely tailored to the specific RF device transmitting the waveform. If another device were to replay the same waveform, it should register as a different unique device due to the differences in the transmitter

To meet these two objectives performers may either learn a new dialect (e.g. a modification) of an existing standard (such as one of those being identified by Task 1A), or a wholly distinct waveform. Proposers should describe how their system will meet the above two objectives. If the proposed approach is to learn a new dialect of an existing protocol, it is expected that the first objective is met by demonstrating that the new dialect is still interpretable by existing devices which implement that protocol. For example, a COTS WiFi device should still be able to interpret a learned WiFi dialect. If the proposed approach will learn a wholly new waveform, proposers should pay careful attention to the need to implement both a transmitter and receiver for this waveform.

Proposers should clearly address how their system traverses the combinatorically infinite space of waveform design choices. A system that merely brute forces its way through waveform design will not converge to a viable solution within a tractable timeframe. Additionally proposers should address what unique aspects of each transmitter they expect their system will learn to exploit.

Performers must provide their own RF hardware for Task 1B. In Phase 1, at least 10 nominally identical RF devices should be provided, for example, SDRs (software-defined radios). The RF systems must be commercially available so that a Government laboratory can purchase identical units for verification testing. Alternatively, additional RF systems can be provided by Task 1B performers to the government team for independent evaluation.

During Phase 2, Task 1B performers will improve the performance of their system and expand the size of the overall population of transmit devices used to at least 100. Larger populations are desirable to the extent practical.

Performance Metrics Each Task 1B system should find a transmit waveform that allows it to be consistently and uniquely distinguished from all other transmitters mimicking the same waveform. Performers must verify that the final waveform selected by each transmitter enables the Task 1A system to consistently recognize that transmitter and distinguish it from all other transmitters in the Task 1A database.

Task 1B performance goals are shown in the following table. Here, PD is defined as the probability that the transmitter can be distinguished from a signal in the Task 1A database.

Description Threshold Goal

Task 1B

A transmitter increases the PD of a Task 1A system

PD > 0.95 with a population of 1k transmitters

PD > 0.99 with a population of 10k transmitters

Additionally, performers choosing to learn a dialect must verify that the dialect maintains the baseline communications data rate. Performers choosing to learn wholly new waveforms should propose a method for verifying that they realize the same data rate per unit bandwidth as the baseline.

TA1 Schedule of Deliverables

Phase 1 Phase 1 Code Delivery 11 MAC* Deliver final algorithms for both training and evaluation to a government laboratory for independent evaluation Phase 1 Final Report 12 MAC Including performance test results

Phase 2 Phase 2 Code Delivery 23 MAC Deliver final algorithms for both training and evaluation to a government laboratory for independent evaluation Real-time demonstration 24 MAC Completed demonstration of Task 1A real-time evaluation using streaming input data from the TA3 RF system. Real-time demonstration of Task 1B transmit data rate and transmitter distinguishability.

Phase 2 Final Report 24 MAC Including performance test results *MAC = months after contract award

b) TA2: Spectrum Awareness

Background Proliferation of spectrum sharing, in which multiple wireless systems use the same RF frequency bands, has underscored a critical capability need for awareness of the signals actually occupying the spectrum to enable enforcement. Presently most spectrum monitoring is only conducted on-demand, after an “interference event” has been reported. The few automated systems, which do exist are typically limited to reporting which frequencies are being transmitted on due to the impracticality of analyzing and identifying numerous signals over wide bandwidths in real time.

To make this wideband real-time analysis practical, TA2 performers will focus on RF systems that are able to autonomously monitor 5GHz of RF bandwidth. TA2 is comprised of two tasks.

Task 2A focuses on learning attention and saliency while monitoring a 500 MHz bandwidth, detecting, characterizing, and labeling important signals. Task 2B requires autonomously reconfiguring a complex RF system to improve monitoring across the full 5GHz of bandwidth.

Traditional RF systems are typically triggered by a detection (a signal above a certain energy level) and each detection is processed through a series of hand-engineered analysis steps. These systems have no attentional mechanism and thus every signal garners the system’s attention equally, and each signal goes through the same processing steps.

In a spectrum sharing setting, most signals encountered by a receiver are unimportant; they are the types of signals one might expect to find given the operating band and geographic location of the receiver. Conversely, the important signals are those that are unexpected. For example, this could be as simple as a communications signal present in a band where only radars are expected.

Or, importance could involve more sophisticated decision making such as the presence of repetitious signals (i.e., a GPS spoofing device), the presence of a particular type of communications signal (such as a UAV command link in a busy ISM band), or myriad other signals which could be considered important or out of the norm.

Compounding this problem, traditional RF systems fail to efficiently exploit all available degrees of freedom because of the complexity of controlling such a system. Instead, the RF system and associated control is simplified. For example, frequencies and spatial directions are often scanned in a sequential order irrespective of the statistics of the operating environment, and parameters of the RF frontend are typically fixed at design time instead of being considered part of the control of the RF system.

The diagram in 3 shows the relationship between the two TA2 tasks as well as the hardware provided by TA3.

Figure 3 – Block diagram showing the interaction of TAs 2 and 3

TA2 proposers must provide a schedule that details the interaction between the development associated with Tasks 2A and 2B. The schedule should address how to achieve parallel development of Task 2B given its dependencies on TA3. If surrogates will be used during early development the plan should be specific as to how it reduces overall program risk.

Task 2A: RF Attention and Saliency

Task Description Under this task, performers must develop algorithms that are capable of learning important vs.

unimportant signals and their associated labels. Different instances of such a system should be able to operate entirely differently on the same dataset, depending on how the system is trained.

As an example, consider a dataset that contains both communications systems and radar systems.

Under one training modality, the system might learn communications signals are important and to report the modulation associated with them. Under a different modality the system might learn radar signals are important and to identify them simply with the name of the radar system.

Due to the complexity of such a system, Task 2A proposers should specifically address which aspects of their system are learned and which are preprogrammed in some way. Additionally, proposers should address how their system will handle the different timescales involved. For example, radar pulses and communications waveforms typically occur on very different timescales. Identifying the type of waveform present may rely on characteristics that evolve over a timescale longer than a waveform. As the focus of Task 2A is attention and saliency, proposers should compare the computational burden of their proposed system to one which is indiscriminant (i.e. has no concept of attention and saliency).

Government Furnished Information (GFI) For training, Task 2A performers will receive multiple GFI datasets. Each training dataset will have the following components:

1. Unlabeled background – consisting of background signals which are assumed to be unimportant, and do not have an associated label

2. Important foreground – consisting of signals which could be considered to be important and their associated labels.

This approach allows TA 2 performers to test the generality of their system. By mixing and matching various datasets the definition of an important signal can be varied, as can the labels and the background. To the degree possible, signals in the dataset will be “open-air,” that is the transmitted signal propagates through a real wireless environment.

Performance Metrics For evaluation, the Task 2A system will be trained and tested using a previously unseen dataset and definition of importance. The performance metrics are , the probability of detection of 𝑃𝐷 important signals, and , the probability of false alarm (unimportant signals erroneously 𝑃𝐹𝐴 flagged as important). The following table defines expected performance. Note that important signals are relatively rare, 10% or less of the total signal count.

Description Threshold Goal Task 2A

A receiver detects, characterizes, and labels important but rare signals

500 MHz bandwidth with 10% important signals, > 0.5 and 𝑃𝐷

< 0.5𝑃𝐹𝐴

500 MHz bandwidth with 1% important signals, > 0.9 and 𝑃𝐷

< 0.5𝑃𝐹𝐴

Task 2B: Autonomous RF Sensor Configuration

Task Description Task 2B will build on the Task 2A system by creating algorithms capable of learning to control a complex RF sensor provided by TA3. For planning purposes performers should expect to control the full gamut of controls specified as desirable in the TA3 description. Task 2B will enable demonstration of Spectrum Awareness across at least 5GHz of RF bandwidth.

Since TA3 systems will not be available at the start of the program, Task 2B proposers should describe what they will use as an interim surrogate. Task 2B proposers should fully describe how they make use of the TA3 system once it is available. Additionally, thorough detail should be provided which describes the differences between the training and evaluation processes for a Task 2B. The number of degrees of freedom and the associated size of the control space provided by a TA3 RF system is likely to be combinatorically unwieldy. Proposers should describe how their approach is able to effectively learn a large control space without resorting to brute force trial-and-error.

Performance Metrics Task 2B controls the data received by the Task 2A system. Task 2B performance will be scored based on Task 2A performance, with expectations defined in the following table.

Description Threshold Goal Task 2B

A system tunes a 500 MHz receiver and detects important but rare signals. Control of other aspects of the RF system may be possible.

5 GHz tunable bandwidth with 1% important signals, PD > 0.5 and PFA < 0.5

5 GHz tunable bandwidth with 1% important signals, PD > 0.9 and PFA < 0.5

TA2 Schedule and Deliverables

Phase 1 Phase 1 Code Delivery 11 MAC Deliver final algorithms for both training and evaluation to a government laboratory for independent evaluation Phase 1 Demonstration 12 MAC Completed Task 2A demonstration of offline learning and online detection and labelling of important signals.

Completed Task 2B demonstration of offline learning and online control of the TA3 RF System.

Phase 1 Final Report 12 MAC Including performance test results Phase 2

Phase 2 Code Delivery 23 MAC Deliver final algorithms for both training and evaluation to a government laboratory for independent evaluation

Real-time demonstration 24 MAC Completed demonstration of real-time evaluation of combined Task 2A and Task 2B system using streaming input data from the TA3 RF system

Phase 2 Final Report 24 MAC Including performance test results Phase 3

Support Integrated Demo 24-36 MAC Support development of an integrated real-time open air demonstration with TA3

Phase 2 Final Report 36 MAC Including performance test results

2. TA3: RF System Integrator and Demonstrator

Technical Area Description TA3 will provide a reconfigurable RF system to be used as follows in the RFMLS program:

1. Dataset generation for TA1 and TA2

2. Real-time streaming data for TA1 and TA2 demonstration

3. Demonstration of the integrated TA2 system

Additionally as the system integrator, TA3 performers will be responsible for planning and executing the outdoor real-time demonstration events.

DARPA does not seek to develop new RF hardware or systems under RFMLS TA3. Instead, the RFMLS program will leverage already existing RF system designs that provide an appropriately modern and complex RF system which can benefit from autonomous reconfiguration, and is appropriate for both the IoT RF Fingerprinting and Spectrum Awareness task areas. Proposals seeking to fund the development of entirely new RF system designs will not be selected.

Proposed designs should have existing baselines that have previously or are concurrently under-going development and test. Minor design variations and new manufacturing for RFMLS are acceptable.

An objective of the RFMLS program is to enable autonomous reconfiguration of a complex RF system. To pursue that objective, DARPA seeks the following characteristics for the RF system provided by TA3:

Reconfigurable instantaneous bandwidth (IBW) Maximum IBW of at least 500MHz Multiple (4-16) phase-coherent receivers Resident digital beamforming Center frequency coverage: 1-6 GHz Aperture(s) appropriately covering this frequency range (for phase 3 demonstration)

These specifications provide the minimum control surface necessary for TA2. Additional reconfigurability in the RF system, which would provide a more complex control space and ultimately a more useful autonomous sensor, is also desirable. For example:

Reconfigurability of the analog RF frontend (prior to A/D conversion) Reconfigurability of the antenna aperture

Making such a system Machine Learning “friendly” requires an appropriate API (Application Programming Interface) or ICD (Interface Control Document) to control all necessary faculties.

Proposers under TA3 should describe how their system is controlled, whether an API/ICD already exists or will be developed, and how frequently TA2 performers can modify the configuration of the system.

The RFMLS program does not seek to achieve any specific form-fit or to minimize SWAP. To that end, TA3 performers should be prepared for TA1 and TA2 performers to use adjunct (e.g.

off-board) processing in addition to, or in-lieu of using available on-board processing. To support this the RF system provided by TA2 must have sufficient ability to exfiltrate raw, or near raw RF data through standard high bandwidth interfaces such as 10/40/100 Gb Ethernet, or other similar high-speed external buses.

Proposers should address specifically what Degrees of Freedom (DoF) their system provides, why they believe the proposed system would benefit from autonomous control, and what the system might ultimately learn that a hand-engineered controller could not learn. Additionally proposers should be sure to describe in detail the availability of the proposed RF system and a description of the third year demonstration to be conducted.

TA3 is comprised of three 12 month phases. TA3 will provide multiple RF systems to TA2 and Government evaluators by 6 MAC in Phase 1. For planning and budgetary purposes TA2 performers should expect to deliver 5 RF systems. The ability to deliver RF systems sooner will reduce overall Program risk. In this Phase 1, TA3 will also work with the government team to integrate the RF system into the government team’s dataset generation architecture to gather additional datasets to be used in Phase 2. During Phase 2, TA3 performers will support TA2 in developing an integrated system. During Phase 3, TA3 performers will be responsible for planning, managing and executing open-air demonstrations of the integrated TA2 system.

Schedule & Deliverables

Phase 1 RF System Delivery 6 MAC Deliver 5 reconfigurable RF Systems Dataset Generation 6-12 MAC Integrate RF System to government dataset generation infrastructure Phase 2

Integrate TA2 12-24 MAC Support TA2 integration and real-time demonstration on TA3 hardware

Phase 3 Demo Plan 24 MAC Deliver a plan for Phase 3 demonstration of the integrated system

Phase 3 Support Integrated Demo 24-36 MAC Support development of an integrated real-time open air demonstration with TA2 performer Final Demonstration Dry Run 33 MAC Dry run of final demonstration Final Demonstration 35 MAC Final demonstration Phase 3 Final Report 36 MAC

II. Award Information

A. General Award Information

Multiple awards are anticipated. The amount of resources made available under this BAA will depend on the quality of the proposals received and the availability of funds.

The Government reserves the right to select for negotiation all, some, one, or none of the proposals received in response to this solicitation, and to make awards without discussions with proposers. The Government also reserves the right to conduct discussions if it is later determined to be necessary. If warranted, portions of resulting awards may be segregated into pre-priced options. Additionally, DARPA reserves the right to accept proposals in their entirety or to select only portions of proposals for award. In the event that DARPA desires to award only portions of a proposal, negotiations may be opened with that proposer. The Government reserves the right to fund proposals in phases with options for continued work at the end of one or more of the phases, as applicable.

Awards under this BAA will be made to proposers on the basis of the evaluation criteria listed in Section V, “Application Review Information”, and program balance to provide overall value to the Government. The Government reserves the right to request any additional, necessary documentation once it makes the award instrument determination. Such additional information may include but is not limited to Representations and Certifications (see Section VI.B.4., “Representations and Certifications”). The Government reserves the right to remove proposers from award consideration should the parties fail to reach agreement on award terms, conditions and cost/price within a reasonable time or the proposer fails to timely provide requested additional information. Proposals identified for negotiation may result in a procurement contract or other transaction, depending upon the nature of the work proposed, the required degree of interaction between parties, whether or not the research is classified as Fundamental Research, and other factors.

Proposers looking for innovative, commercial-like contractual arrangements are encouraged to consider requesting Other Transactions. To understand the flexibility and options associated with Other Transactions, consult www.darpa.mil/work-with-us/contract-management#OtherTransactions.

In all cases, the Government contracting officer shall have sole discretion to select award instrument type, regardless of instrument type proposed, and to negotiate all instrument terms and conditions with selectees. DARPA will apply publication or other restrictions, as necessary, if it determines that the research resulting from the proposed effort will present a high likelihood of disclosing performance characteristics of military systems or manufacturing technologies that are unique and critical to defense. Any award resulting from such a determination will include a requirement for DARPA permission before publishing any information or results on the program. For more information on publication restrictions, see the section below on Fundamental Research.

http://www.darpa.mil/work-with-us/contract-management#OtherTransactions

B. Fundamental Research

It is DoD policy that the publication of products of fundamental research will remain unrestricted to the maximum extent possible. National Security Decision Directive (NSDD) 189 defines fundamental research as follows:

‘Fundamental research’ means basic and applied research in science and engineering, the results of which ordinarily are published and shared broadly within the scientific community, as distinguished from proprietary research and from industrial development, design, production, and product utilization, the results of which ordinarily are restricted for proprietary or national security reasons.

As of the date of publication of this BAA, the Government expects that program goals as described herein may be met by proposers intending to perform fundamental research and proposers not intending to perform fundamental research or the proposed research may present a high likelihood of disclosing performance characteristics of military systems or manufacturing technologies that are unique and critical to defense. Based on the nature of the performer and the nature of the work, the Government anticipates that some awards will include restrictions on the resultant research that will require the awardee to seek DARPA permission before publishing any information or results relative to the program.

Proposers should indicate in their proposal whether they believe the scope of the research included in their proposal is fundamental or not. While proposers should clearly explain the intended results of their research, the Government shall have sole discretion to select award instrument type and to negotiate all instrument terms and conditions with selectees. Appropriate clauses will be included in resultant awards for non-fundamental research to prescribe publication requirements and other restrictions, as appropriate. This clause can be found at www.darpa.mil/work-with-us/additional-baa.

For certain research projects, it may be possible that although the research being performed by the awardee is restricted research, a subawardee may be conducting fundamental research. In those cases, it is the awardee’s responsibility to explain in their proposal why its subawardee’s effort is fundamental research

III. Eligibility Information

All responsible sources capable of satisfying the Government's needs may submit a proposal that shall be considered by DARPA.

A. Eligible Applicants

1. Federally Funded Research and Development Centers (FFRDCs) and Government Entities

a) FFRDCs http://www.darpa.mil/work-with-us/additional-baa

FFRDCs are subject to applicable direct competition limitations and cannot propose to this BAA in any capacity unless they meet the following conditions: (1) FFRDCs must clearly demonstrate that the proposed work is not otherwise available from the private sector. (2) FFRDCs must provide a letter on official letterhead from their sponsoring organization citing the specific authority establishing their eligibility to propose to Government solicitations and compete with industry, and their compliance with the associated FFRDC sponsor agreement’s terms and conditions. This information is required for FFRDCs proposing to be awardees or subawardees.

b) Government Entities

Government Entities (e.g., Government/National laboratories, military educational institutions, etc.) are subject to applicable direct competition limitations. Government entities must clearly demonstrate that the work is not otherwise available from the private sector and provide written documentation citing the specific statutory authority and contractual authority, if relevant, establishing their ability to propose to Government solicitations.

c) Authority and Eligibility

At the present time, DARPA does not consider 15 U.S.C. § 3710a to be sufficient legal authority to show eligibility. While 10 U.S.C.§ 2539b may be the appropriate statutory starting point for some entities, specific supporting regulatory guidance, together with evidence of agency approval, will still be required to fully establish eligibility. DARPA will consider FFRDC and Government entity eligibility submissions on a case-by-case basis; however, the burden to prove eligibility for all team members rests solely with the proposer.

2. Non-U.S. Organizations and/or Individuals

Non-U.S. organizations and/or individuals may participate to the extent that such participants comply with any necessary nondisclosure agreements, security regulations, export control laws, and other governing statutes applicable under the circumstances.

B. Organizational Conflicts of Interest

FAR 9.5 Requirements In accordance with FAR 9.5, proposers are required to…