From cell phones to self-driving cars: processing the signals that connect us
Engineering was always his dream, but ending up in the field of signal processing was a result of the times for Anum Ali.
Ali completed his undergraduate degree in 2011 just as the telecommunication boom was kicking in. Engineering offered the perfect balance of mathematics and application, and signal processing became the most viable way for him to pursue his passion.
Now a doctoral candidate at the The University of Texas at Austin, Ali’s research tackles the future of wireless communication.
“In communications, we transmit something and by the time it gets to the receiver, it’s corrupted by different contaminants,” Ali explained. “Signal processing is about taking that impure signal and extracting meaningful information from it.”
Newer tech requires higher and higher rates of data transmission — rates that our existing frequencies can’t provide.
It’s an area rich in mathematics, which means most of Ali’s day-to-day work consists of reading the latest papers, learning techniques and theorizing how those new methods could help answer his questions.
And what are Ali’s questions, exactly?
“Most wireless communication today is on frequencies below 6 GHz,” Ali explained. Wi-fi, for example, operates at 2.4 GHz and 5 GHz while cell phones use 600 MHz to 2.5 GHz. “The problem is that all these frequencies are now saturated.”
As technology advances, this issue becomes increasingly necessary to solve. Newer tech requires higher and higher rates of data transmission — rates that our existing frequencies can’t provide.
“As we move into the next generations of wireless communication, we will be connecting more than people… Cars will be talking to other cars and to the infrastructure.”
“It may seem very obvious that, if it were possible to move to higher, unused frequencies, why haven’t people done it so far?” Ali said. “What happens is that when we move on to higher frequencies and transmit a certain power, the amount we receive decreases very rapidly.”
The decay is so severe that if the same power is transmitted at 2 GHz and 20 GHz, the power received at 20 GHz is only one percent of the power received at 2 GHz.
“In our research group, we look into the problems of how even at these frequencies, when you receive very low signal powers, you could still be able to communicate reliably.”
Frequencies in this higher range, between 30 GHz to 300 GHz are called millimeter waves. To support these frequencies, multiple antennae at both the transmitting and receiving ends are needed. But that adds signal processing challenges.
“From the transmitter to receiver is a wireless channel, which must be learned to be able to communicate on it,” Ali explained. This is achieved through a process is called “training.”
“When you increase the number of antennae to 10s or 100s, the training period becomes so large that essentially you’d just train all the time and have no time left for sending data.”
At lower frequencies, learning channels is an established process. In a cell phone to cell phone call, for example, both phones “learn” the channels from device to station and thus establish a connection.
“What I try to do is take the channels at low frequencies, which are very easy to learn, and then extract information about millimeter wave channels so as to reduce the training required,” Ali said. “The level of correlation in these two bands is unclear, so that also poses another challenge.”
It’s an exciting challenge for Ali, who envisions a future where signal processing will mean more than Wi-Fi routers and mobile calls.
“As we move into the next generations of wireless communication, we will be connecting more than people,” he said. “One of the things we’ll be connecting is cars. Cars will be talking to other cars and to the infrastructure. This communication is called V2X — vehicular to everything — communication.”
This means that in lieu of traffic signals, cars will learn to discuss amongst themselves who goes first. Their discussion, however, will inevitably generate large amounts of data that will need higher frequencies to transmit on.
“As the cars move, the channel will change and you have to learn it repeatedly. But cars move very fast, so the rate at which you have to learn the channel, again and again, is really high. So there comes my work: you use some low frequency channel to make sure you can learn these dynamic, highly time varying channels.”
In a field with such obvious real-world applications, it can nevertheless be tricky keeping code open-source. Ali estimated that half of the projects in his field are in conjunction with industry partners who request non-disclosure agreements.
At the same time, Ali said that using online platforms to showcase and share code is becoming more of a need than an option for researchers.
“If your fellow scientists can’t reproduce your work, it could discredit your research. Non-reproducible research hinders the progress of the entire field.”
“As we’re moving away from fundamental research, the types of simulations that researchers are doing are very involved and it’s becoming practically impossible to write all the simulation parameters in a 14-page paper,” he said. “In order to fit their content into the page limit given by the publication, they have to cut more and more from the details of the simulation and this makes it practically impossible to reproduce anyone’s work.”
If your fellow scientists can’t reproduce your work, Ali added, it could discredit your research. In the same vein, non-reproducible research hinders the progress of the entire field.
“I think open source is a move in the right direction… If someone has already solved a problem, you should not be wasting a year trying to reproduce that work!” Ali said. “From a selfish point of view also, I believe it can attract more interest to your work.”
Ali’s best advice for those seeking to share complex simulations and ensure reproducibility is, unsurprisingly, rooted in communication too: Phone a friend.
“Once you’ve cleaned your code for upload, it is helpful to ask your colleagues who aren’t aware of your work to regenerate a figure. If he can regenerate it easily, most likely other people can too. If he cannot, then, of course, you need to put more effort into cleaning it and making it accessible to other people.”
Ali’s first foray into open source science came from his Master’s advisor, Professor Tareq al-Naffouri at King Abdullah University in Saudi Arabia.
“You are only as good as your teacher, so think about your mentors and advisors very carefully.”
“We started doing it using our own website. Now, my Ph.D. advisor, Professor Robert Heath, is also encouraging students to make code available online, primarily using Code Ocean and GitHub. They’ve both had an impact on me.”
“At every stage, there’s somebody who drives you or pushes you,” Ali added. “It’s very important to find the right mentor. To have somebody whose thinking process is clearer than yours, so you see how they think about problems. You are only as good as your teacher, so choose your mentors and advisors very carefully.”
With only a year left until his doctorate is complete, Ali is looking forward to the next chapter. Working at an established company is one promising path, but he also aspires to create a start-up of his own someday — one focused on applying machine learning to wireless communication.
“It’s something a lot of people will be very interested in, in the next 10 years!”
While talking cars may sound like science fiction, Ali’s field is working on getting us there.
Originally published at codeocean.com.