In this interview, we talk to Scott Kuzdeba, author of the book Radio Frequency Machine Learning: A Practical Deep Learning Perspective. We discuss the motivation behind writing the book, the target audience, the most useful aspects of the book, the challenges of writing the book, and advice for other engineers who are considering writing a book.
Scott Kuzdeba has been leading the development of RF machine learning (ML) algorithms and systems for over 15 years. He has served in numerous roles including chief scientist, principal investigator, subject matter expert, and capture leader. Currently Dr. Kuzdeba is the director of the Edge and Spectral Artificial Intelligence group within the BAE Systems FAST Labs™ research and development organization. In this role, he works closely with governmental organizations (including DARPA, IARPA, and AFRL) and multiple university partners to develop new RF ML solutions to solve some of the defense industry’s most challenging problems. Dr. Kuzdeba frequently publishes peer reviewed papers, holds several patents within the RF ML field, and is an active member of the AOC, AAAI, and IEEE. He received his PhD in computational neuroscience from Boston University, where the inspiration for his artificial intelligence work stemmed from studying biological neural networks related to speech perception and articulation
1. Could you summarize the main content of your book? What are the key topics addressed?
The book covers how machine learning (ML) can be applied to radio frequency (RF) applications, with a primary focus on how to use deep learning (the key technology behind the current wave of artificial intelligence (AI)). The book approaches these discussions from two different angles: 1) from the ML field and how problems and techniques are discussed in the wider AI field and 2) from traditional approaches to signal processing and RF system development on how AI fits within, and can make use of, the existing best practices. The ML focus breaks down the discussion in terms of supervised learning (classification), unsupervised learning (clustering), reinforcement learning (sensor control), and the hot subfield of generative AI (waveform synthesis). The more RF focused discussions look at how AI fits within a complete RF system, how to design symbiotic traditional digital signal processing and AI solutions, how to mitigate and design for the dynamic and complex RF environments and scenarios the system may eventually face, as well as how to deploy solutions within edge hardware (e.g., consumer electronics). Throughout, there is dedicated discussions on the crucial need for data that enables the success in developing and training all these approaches.
2. What is the primary purpose of your book? How do you envision it helping readers in their work or studies?
The primary purpose of this book is to help practitioners break into the emerging field of RF ML or become more comfortable in leveraging the successes of other AI domains and determining how to modify and design RF-native solutions. The book was written to try to gradually build an intuitive understanding of the vast directions that the AI field is going and ensure the reader can develop a strong core understanding of the field to be knowledgeable in discussions, development, evaluations, and personal growth. The book can be read from cover to cover to systematically walk through AI techniques and how they can be applied to RF systems or readers can bounce in and out of sections to cover relevant topics, as each chapter was written to stand on its own. Those seeking a deeper theoretical knowledge of AI can read this book in parallel with deeper dives into the vast and well-articulated AI literature (with many references included within the book), creating concrete mappings between what is discussed in other domains (e.g., computer vision) and its relevance and gaps for the RF domain. Similarly, those seeking implementation practice can read this book while following many of the great online coding tutorials (references also included), helping to connect the practical coding implementations with a deeper understanding and wisdom of what is happening.
3. What sets your book apart from other works in the same field? Are there any innovative concepts, methodologies, or insights that make it stand out?
The RF ML field is new, and as such, there are not many books on this topic quite yet. Those that do exist are largely focused on diving into deeper theoretical underpinnings (which limits the number of people in the field that are ready to dive in), have different chapters written by different authors and are therefore not as cohesive and comprehensive, or are longer journal articles and the like that are typically more in the style of a survey paper. This book is instead focused on providing breadth across the RF ML field through cohesive coverage of topics and common grounding examples. Further, this book is aimed at making the field intuitive and easier to understand for readers who are at various levels in the RF ML field, from those that are just starting out, to those that are looking to add to their knowledge or seek thought provoking ideas. Insights are shared throughout the book that help make the content easily consumable and readily understandable within the context of the RF field.
4.Who is the intended readership for your book? Are there specific industries, professionals, or fields of study that would benefit most from this content?
This book is intended for anyone interested in RF ML, from those just starting out to those who have been working in the field for years. The target audience are those who either work within, or are learning about, the RF domain and would like to understand how AI plays a role. It is therefore assumed that readers already have some prior knowledge of RF. While understanding of AI is not required, some prior exposure would be helpful. The book is written to target practitioners, but likely is a good academia supplemental resource as well. The book speaks across these audience groups by providing high level concepts, examples going a bit deeper, and jumping off references for further reading, while aiming to stimulate thoughts throughout.
5. What are the most important lessons or insights you want readers to take away from this book?
The most important thing to understand from this book is that RF ML is here, it will only expand, and it is not as simple as “plug and play” from other domains (e.g., you cannot pick up your favorite large language model (LLM) (i.e., ChatGPT) and naively apply it to RF signals). This book will help readers become comfortable in this new, emerging area, providing them with the tools to think critically about how the current wave of AI will impact our RF systems.
6.Does your book include any original research, case studies, or data? If so, could you highlight some of the most significant findings?
A lot of the examples provided in the book are from my own research, which include citations to more detailed research papers. My team and I have been conducting research in this area and have spent a lot of time looking into the nuances, differences, and open research questions in applying deep learning to the RF domain. I share many of these throughout the book. This includes using common deep learning architectures for grounding the discussion and helping to compare/contrast how things change as applications, data, and use cases change. This is likely most seen throughout the book in the conversations of the deep learning feature extraction part of the networks and how common core elements get repurposed for the different types of learning (supervised, unsupervised, reinforcement, generative) or within different RF contexts (deployed at the edge, integrated in with traditional digital signal processing, designed to be more robust to environmental unknowns, etc.). As for data, data is key to the solutions discussed throughout the book, with an entire chapter devoted to discussing data (with many references to data repositories, tools, and resources).
7.Does your book address any new or emerging trends in the field? How does it prepare readers for future developments?
I would argue that the RF ML field itself is an emerging field. Many of the techniques (deep learning, and more broadly machine learning) are not new and have a history in the RF domain. However, the scale and pace at which they are being applied and matured across domains is new, as we have seen across domains and industries. As an example, when I started writing this book, ChatGPT didn’t exist and most people where not familiar with large language models (a term that would come later) or the interesting research being conducted using transformer architectures. Limited work had been done on applying transformers to the RF domain and I had included a brief blurb about it, but as time progressed, I had to update parts of the book to point to the potential of this new technology. Beyond LLMs, I point to several intriguing future developments throughout the book, including providing some thought-provoking directions at the end of the Introduction.
8.What personal experiences, if any, have shaped your perspective or approach to the topics discussed in your book?
My academic career had heavy theoretical engineering, mathematics, statistics, and science components to it. I grasped complex concepts through rigorous dedication and practice, with intuition often not coming until the end. In parallel to this, I picked up many machine learning concepts through various different sources (self-taught, interaction with others, formal settings). One of the things that I started to see in some AI/ML resources and interactions was a trend towards making concepts first intuitive and easily understood and only afterwards diving deeper into their theoretical underpinnings. I found this progression to be best for diving into something quickly and feeling comfortable while doing it. I have tried to emulate this mindset in this book, making it easy for others to gain intuition about RF ML while gaining breadth and a deeper understanding of the field.
Learn more about the book on our websites:
ARTECH HOUSE USA : Radio Frequency Machine Learning: A Practical Deep Learning Perspective
ARTECH HOUSE U.K.: Radio Frequency Machine Learning: A Practical Deep Learning Perspective
More Microwave and RF content here: Microwave and RF – Artech House Insider
