One of the most satisfying aspects of working at Cisco is tackling a thorny problem of engineering or computational science, and seeing your work reflected in a global standard or a shipping product within a few years. In wireless networking in particular, we have teams on the leading edge of research, software and hardware engineering, and on applying AI to wireless systems. There are many interesting problems our engineers have solved.

But there are many we still haven’t.

I was recently asked to present to a National Science Foundation workshop on “Future Wireless Research Challenges.” The NSF knows that successful research leads to great benefits. It provides funding to academic institutions to work on promising research areas, but predicting what research areas will lead to significant breakthroughs is extremely difficult. Bringing together the collective intelligence of experts from both industry and academia improves the odds.

Below is a simplified view of my recommendations to the NSF workshop.

Using New Spectrum Wisely

We have outgrown the radio spectrum we use for wireless communication. Fortunately, in 2020 we will see growth in the use of new frequency bands. First, we’re going to see high frequencies, from about 30 to about 60 GHz, called “millimeter wave” rolled out for mobile data communications (5G) and as an extension of Wi-Fi (802.11ay).

Signals in these short wavelengths can be pointed by antenna arrays with high directionality, which improves spectrum use by limiting interference between devices. There’s a challenge in improving how radios do signal acquisition and tracking for these focused beams. Assuming we succeed at optimizing our use of mmWave, the next band of terahertz frequencies provides even greater directionality.

Second, some frequency bands in the 3.5 GHz range (close to existing cellular bands), that have been traditionally been off-limits, will be slowly opened up for private and shared use in the U.S. Use of these bands is contingent on effective sharing of their frequencies with the incumbent users of them, as well as relinquishing them when necessary.

With all the new spectrum coming online as well as an explosion of new devices, a new challenge for radios will be selecting which frequency band to use of the multiple available; and how large numbers of diverse devices and multiple flavors of wireless infrastructure will communicate and negotiate spectrum use among themselves.  There can be huge value if wireless devices and infrastructure get better at dynamically and intelligently sharing the precious wireless spectrum. This corresponds to the two-decades old vision of cognitive radio, which is slowly coming closer to reality.

Maximizing Performance

Wireless today is very fast, but performance falls off considerably in complex or crowded environments. If a sporting arena wanted to provide every person in every seat with their own customized augmented-reality experience, we couldn’t do it efficiently today. (However, there are creative proposals, such as putting a wireless access point under every attendee’s seat, which is both cost-prohibitive and would lead to interference issues.)

New wireless standards will help.Wi-Fi 6, for example, provides significant benefits over prior generations of Wi-Fi in terms of predictability and aggregate throughput that scales gracefully as the number of devices increases.   This is thanks to scheduled transmission (vs. listen-before-talk in prior versions of Wi-Fi), and the use of OFDMA for more efficient spectrum use. (Wi-Fi 6 and 5G use fundamentally the same radio modulation and signaling constructs, but optimized for different use cases.)

Wi-Fi 6 scales up much more gracefully than previous versions of the standard.

But it is still challenging to design radio systems that can take full advantage of these new protocols. In Wi-Fi 6, for example, access points and devices can use several methods to optimize their use of spectrum. The challenge is to get them to select and jointly perform the best optimization. For example, a single wireless access point needs to determine, in real-time, for all the devices in its coverage, what is the best set of packets (to a single or multiple receivers) to transmit next – how to encode it and schedule it – while the wireless environment is rapidly changing along with the traffic demands on it.

Furthermore, each access point likely exists in an environment with other APs. It is challenging, to put it mildly, to get all of the APs and devices to optimize their spectrum use to maximize the quality of experience (QoE) for every application in use, continuously, in real-time, and with low complexity (for low cost). One of the engineering challenges is that it is not the average or median performance that matters, but rather the tail of the probability density function: We want the tail to be as small as possible, since even one late packet out of 10,000 may degrade application performance.  There are a variety of promising research directions to address this, including machine learning which learns patterns associated with good schedules, and various forms of scheduling optimization that leverage structural properties of the ideal solution.

Networks that Know Themselves

In a few years, literally billions of new wireless devices will come online. How should each wireless network treat each new device as it connects? Some devices will need lots of bandwidth. Some will require ultra-low latency. Some will be battery-power-limited. Some may be malicious.

It is surprisingly challenging for network admins to know what is even on their networks. Not all devices identify themselves. Malicious devices can lie about what they are. Furthermore, as more traffic becomes encrypted (as it should be for security) identification becomes even harder.  But machine learning can fingerprint devices based on how they use the network (for example, who they talk to, frequency of transmission, or size of packets). And once a device or application has been classified as a particular type, analytics can be applied to confirm that the device continues to behave as expected, or to determine that it is malfunctioning, or that it has been compromised and is a security threat.  There are significant research opportunities to improve security simply by allowing networks to know what devices are on them.

What’s a blip and what’s a problem? Machine learning should be able to tell the difference.

The growing complexity will also make it more difficult to know if a network is operating as intended or if there is a problem. Ideally, for speed of reaction and for scalability, the network itself should be able to determine this, using AI if necessary, so it can alert an IT staffer when there’s an issue outside the bounds of expected or desired performance.

In our work so far, machine learning, fed by huge amounts of rich, contextual information, has allowed  us to reduce the alerts a network sends to its operator. In some situations, the reduction is one or two orders of magnitude.  These are huge improvements, but there are opportunities for more.

In particular, once problems are found, we need to resolve them quickly. We believe machine reasoning is a key capability to identify the root causes of problems, and to identify likely fixes. This will help turn the dream of the “self-healing” wireless network into reality.

Network as Sensor

We can use wireless networks for more than just data transfer. Network devices are constantly painting their environments with radio waves; how those waves are reflected back provides useful information about the environment.

Today, we can use various techniques to geolocate devices indoors, where GPS doesn’t work: We collect data based on received signal strength, time-of-flight, and angle-of-arrival to estimate the location of various devices relative to indoor APs. Improving the accuracy, frequency, and scale of estimating indoor locations can open up many applications, like autonomous indoor robots.

We can also use wireless transmitters for “RF imaging” of the environment.  A wireless AP can transmit a signal and listen for the response. From the time-varying multipath response received we can learn about the environment. Is there movement? How many objects? Where are they?  How fast are they moving? This could even also be used to determine a person’s breathing rate (periodic changes in a person’s chest lead to periodic changes in the multipath) and other attributes.

Compared to using optical imaging with a video camera, RF imaging provides several benefits including a greater degree of privacy because the identity of the individual is not known, and increased utility of access points. It is a promising area with potentially many surprises.  We should investigate which frequencies are best for RF imaging of an environment, and how much can be accomplished using communication-like signals, rather than a more radar-like approach. We should also learn more about how can we fuse information from multiple frequency bands and multiple APs together.

Radios as Software

There are fascinating engineering challenges ahead in the field of  software-defined radios (SDRs) and cognitive radio . Rather than having to engineer all the advancements I discussed above into new radio hardware, and worry about them quickly becoming outdated, we may be able to add capabilities to our wireless systems through software updates – just like we update our smartphones today. As new spectrum or encoding methods become available, we could update existing devices in place.

SDR technology has yet to achieve the low power, high performance, and low cost of traditional hardware radios, but advances in Moore’s Law, flexibility in designing processor architectures (like RISC V), and other areas are making SDRs more economically practical.  Software-defined radio will lead to profound opportunities in how we design and operate wireless networks and devices.


I have covered some of the most promising opportunities in which we can advance wireless technology. It is great to see NSF funding more research in the field. You can learn more in the published research in IEEE and ACM: Do an Internet search using selected keywords from this post plus either “IEEE” or “ACM” and you will get pointers to publications in the digital libraries, paper archive sites, and professors’ websites.

I hope this information inspires you to learn more about these areas, create innovative solutions, and build something great.

Learn more about our current wireless products: Cisco Wireless and Mobility.




John Apostolopoulos

Vice President & Chief Technology Officer

Intent Based Networking Group & Innovation Labs