People of ACM - Andrea Goldsmith

October 2, 2018

You have been in the wireless technologies field since the 1980s. How did you become interested in pursuing a career in engineering, and specifically, wireless computer communications networks?

I didn’t start college thinking I would become an engineer. My interests as a teenager were very broad, shaped in part by the influences of my two parents; my dad was a mechanical engineering professor and my mom was a cartoonist and character designer for the Rocky and Bullwinkle show (in fact, the character Natasha Fatale was modeled after her). I also lived in Europe the year before starting college, which sparked a passion for foreign languages and political science. While I enjoyed math and science in high school, I didn’t find them to be particularly captivating. My dad suggested I declare engineering on my college applications since it would be easier to switch into the humanities than the other way around. This seemed like wise advice, so I followed it.

My first year at the University of California, Berkeley was hard. I had not taken all the prerequisites for my math and physics classes, thinking I could just pick up whatever I didn’t know along with learning new material. In addition, the environment was very competitive and the quarter system provided little room for error in time management, which was particularly challenging as I was working full time in addition to taking classes. There was also a lot of subtle bias about women not being cut out to do math or engineering, particularly those like me who were not top students in these classes. During the summer between my freshman and sophomore years, I reflected on whether I should continue with engineering or switch over to a political science major. I decided I should give myself another year to decide, while learning from my successes and failures the previous year.

What turned the tide for me during my sophomore year was a math class with an inspiring female teaching assistant, and a somewhat boring required class in European politics. I decided that I did really like math, and engineering would allow me to apply mathematics to solve problems that could benefit people. Hence I decided to major in engineering math. Toward the end of my undergraduate degree, I focused on communications because I liked the application of its rigorous mathematics to engineering, and the aspect of developing a technology that significantly impacted people’s lives. However, I didn’t know if I wanted to be an engineer long term. In fact, I had no idea what engineers actually did.

I decided to work following my undergraduate degree while I figured out my professional ambitions. This was in 1986; the first cellular system had just been launched in Chicago, and Wi-Fi was still years away. Since commercial wireless technology was still in its infancy, there weren’t really any engineering jobs in that area, especially for new graduates. I was fortunate to interview with a small defense communications startup, Maxim Technologies, that was working on multiple-antenna beamforming techniques and satellite communications. The company was very poorly run, and it generally only hired advanced PhDs and engineers right out of college like me. As a result, our group of newly-minted engineers had sole responsibility for solving really challenging problems, many of which were above our heads to solve. I often found myself at Stanford’s engineering library researching problems we were trying to solve. While I was captivated by the elegance and rigor of the papers I read, I didn’t have the technical background to understand them at a deep level. I also found that when I discussed technical problems with the PhD engineers at Maxim, they approached problem formulation and solution through a completely different way of thinking than I was capable of.

I realized that if I wanted to solve hard communication problems, particularly those in the emerging cellular and Wi-Fi systems, I would need to enhance my technical knowledge. So in 1989 I decided to return to graduate school for a Master’s. I initially applied to only two schools, UC Berkeley and Stanford, thinking that if both rejected me I would apply more broadly the subsequent year. Stanford rejected me but Pravin Varaiya, who was in charge of EE graduate admissions at UC Berkeley, saw something special about my application. He admitted me to Cal and to his research group even though there were surely many applicants with better grades and test scores than I had. Pravin was an inspiring adviser, teaching me how to formulate research questions and how to answer them with both mathematical depth and insight. I felt so privileged to be working with him, and was having so much fun doing research, that going on beyond the Master’s for a PhD was a seamless decision.

Beginning in the 1990s, you developed fundamental capacity limits for wireless networks. Will you tell us what the wireless landscape looked like then, and why establishing capacity limits was important?

I began my doctoral research in 1989. It was an exciting time to be working in wireless, as the first generation of cellular technology was experiencing exponential growth, the best technology for the 2G cellular standard was being hotly debated, and the precursors to Wi-Fi systems were starting to emerge. I had already fallen in love with wireless communications during my three years at Maxim. In my doctorate I was intrigued to study the fundamental capacity limits for these emerging wireless systems, and the designs needed to achieve performance close to these limits. Shannon theory was a beautiful and entrancing topic, while also providing practical design insights. How these practical designs performed relative to the Shannon limits provided insights into refining and expanding the wireless information theory problems I was investigating. This research trajectory between the Shannon limits of wireless systems and designs to achieve those limits was far more satisfying for me than focusing on only one of those two research dimensions.

My most satisfying result as a PhD student was my derivation of the capacity for time-varying wireless channels with perfect transmitter and receiver channel state information, and my companion result on achieving close to this limit using adaptive M-ary quadrature amplitude modulation (MQAM). At the time there was no modulation that adapted the data rate to the instantaneous channel signal-to-noise ratio (SNR) in any wireless system, yet Shannon theory made it obvious that this was the right approach to maximize the data rate over fading channels. I was also inspired by the work of John Cioffi and his research group using adaptive modulation on each subcarrier in orthogonal frequency-division multiplexing (OFDM) for twisted pair copper wire channels, which eventually became the asymmetric digital subscriber line (ADSL) standard. Today all Wi-Fi and cellular systems use adaptive MQAM modulation, which I had proposed and analyzed as part of my doctoral research, yet I never could have foreseen that back in the early 1990s, when Wi-Fi was in its infancy and 2G cellular standards were focused on improving quality and user capacity for fixed-rate voice calls.

In a recent talk, you discussed the concept of software-defined wireless networking as an approach that might put various wireless technologies (Wi-Fi, cellular, millimeter wave, cognitive radio) in a seamless wireless cloud to deliver the right performance, energy consumption, latency, etc. based on the needs of the specific device being used. Can you talk a little more about how software-defined wireless might help us meet the challenges of 5G networks and the internet of things (IoT)?

The basic architecture of a cellular system has not changed since it was originally proposed in a 1948 paper by D. H. Ring. With the advances in wireless technology as well as in optimization and machine learning, it is time to rethink cellular system architectures to put real-time adaptation of the network in the cloud. Under this premise, software in the cloud collects measurements from the cellular network base stations and users, which are then used to maximize system performance through adapting parameters such as frequency assignment, transmit power, antenna parameters, and data rates. The same premise can be used for Wi-Fi, which is the underlying premise of Plume’s cloud software. If cloud control is a good idea for cellular and Wi-Fi, then perhaps we should control all wireless networks in the cloud. This would enable resource allocation and application mapping to resources across different wireless networks in an optimal way.

This notion of cloud control for all wireless networks is what I call software-defined wireless networking (SDWN). In this architecture, the different types of radios associated with different networks and frequency bands in the overall system, such as Wi-Fi, cellular, and millimeter wave, are an inexpensive commodity running intelligent software. The centralized controller not only optimizes the radio parameters, but it also maps the applications to the different radios and their networks that are best suited for that application’s requirements. For example, in a low-power IoT application, the SDWN controller takes the most energy efficient network and maps the device to that network. On the other hand, the SDWN controller might assign a virtual reality application to its millimeter wave network as it has plentiful spectrum. For an autonomous driving application, the available network with the highest reliability and lowest latency might be used. In general the SDWN controller will match the application and the device to the network it is best suited to at a given point in time, which will depend on the available bandwidth, energy constraints, proximity, congestion and propagation conditions of the network.

There are a lot of technical and non-technical challenges for SDWNs, but this is a promising vision for a seamless cloud of connectivity. Users do not care which network they are using—they just want it to work and be inexpensive or free. An SDWN architecture could ensure connectivity that meets the requirements of each application if it is possible to do so with all the available network resources optimally allocated. Challenges of this vision include the complexity in optimization of network resources, seamless handover between different networks, and controlling heterogeneous wireless hardware across different networks as well as frequency bands from the cloud.

From your experiences with Plume Wi-Fi and Quantenna, what advice you would offer someone in an academic research setting who is starting a company to bring their invention to industry?

Startups are a wild ride that is not for the faint of heart. The advice I offer academics thinking to start a company is to first consider why they want to do a startup. Since a successful startup exit is rare, I believe academic entrepreneurs should be motivated to do a startup primarily if they are passionate about turning one of their ideas into a technology and then building a company around that technology. They should, along with their family, also be prepared for the time commitment as well as the trials and tribulations of a startup. There are many important decisions to make such as whether to be the sole founder or recruit co-founders, whether to take venture money or bootstrap, and how much time to develop the technology within the university before launching the company. Often what may seem like obvious answers to those questions can come back to haunt a founder years later. So it is important to solicit advice from trusted and experienced sources for each decision fork in the road.

In looking back on my two startup experiences, they provided some of the most rewarding, and some of the most difficult, moments of my professional career. I launched my startups to see if my academic research could form the foundation of a product and company that would have a big impact on the wireless industry. In that regard, I view my startup experiences as successes. The knowledge I gained from each startup enhanced my research and teaching when I returned to the university. I also came to appreciate what a great job academics have in terms of our freedom and the brilliant motivated students we work with. I discovered that earning respect as an entrepreneur is not easy for professors as we are often viewed as “too academic.” My startups taught me that the most important ingredients in a company’s success are its people and culture. I found each startup, and each challenge in them, to be different, which is what makes startups so exhilarating.

In short, I suggest you first make sure you are realistic about what you are getting yourself into. If it still sounds alluring, then go for it, as the experience of doing a startup is a grand adventure unlike any other professional endeavor.

Andrea Goldsmith is the Stephen Harris Professor in the School of Engineering at Stanford University. Earlier in her career, she co-founded and served as Chief Technical Officer of Plume Wi-Fi and Quantenna, and she currently serves on the corporate or technical advisory boards of Crown Castle Inc., Interdigital Corp., Sequans, Quantenna and Cohere. Goldsmith has published 160 journal papers, 320 peer-reviewed conference papers, and three textbooks, one of which has been translated into Chinese, Japanese and Russian. She also holds 29 patents, with three pending.

She has received several awards for her work, including the IEEE Eric E. Sumner Award in Communications as well as membership in the National Academy of Engineering and American Academy of Arts and Sciences. Goldsmith was recently named the 2018-2019 ACM Athena Lecturer for contributions to the theory and practice of adaptive wireless communications, and for the successful transfer of research to commercial technology.