"Our Ultimate Goal is to Save Lives" - NPS Makes AtomicSense of the Radar Scene - CEO Behrooz Rezvani

Neural Propulsion Systems (NPS) uses mathematical breakthroughs to make radar see farther, clearer and faster. Founder and CEO, Dr Behrooz Rezvani, explains the NPS AtomicSense algorithm based on Atomic Norm created by Caltech and MIT scientists.

Rezvani previously founded telecommunications industry companies Ikanos (acquired by Qualcomm) and Quantenna (acquired by ON Semiconductor Corp). Earlier in his career, he worked on MRI imaging systems at GE and high-resolution ultrasound at Siemens Office.

About MRI and ultrasound, he says: "Both systems have fundamental issues detecting signals. In some ways, radar is another system that sends a type of signal and then listens to it."

He started NPS five years ago to find ways for search and rescue drones to find people and animals in tough spots in the woods during fires which could not be seen visually.

"I realized that the mathematics behind such ideas is quite complex. So, I ended up connecting with Professor Babak Hassibi at Caltech, to apply to the latest breakthroughs in mathematics to the radar problem. Together we developed a new paradigm for radar signal processing based on the Atomic Norm" says Rezvani.

Hassibi confirmed that applying Atomic Norm to radar is possible.

The Advantages of AtomicSense Processing

Radar signal processing is needed when the signal comes back because, when it comes back, it is all mushed up with a lot of noise over it, Rezvani says.

"You have to do a lot of digging and extracting the signals. It is no different than what we do in MRI or what we do in ultrasound. Signals coming back are not prepared to be understood until we remove all the unnecessary noises. That has been the tradition with radars." 

With traditional radar, they send a sequence of signatures and then wait for the signature to come back. Then they look into the details to figure out what kind of target and what kind of object created that signature and they try to decipher the signatures. Then they come up with ten or five signatures of the different objects or targets seen, he explains.

"What we do is fundamentally different and is based on joint spatio-temporal processing and new mathematics. The question asked is, 'How many signatures could explain the received image?' Then we look for the most succinct description of the image in terms of these signatures. Essentially, we solve the problem from the other side." 

An analogy of NPS AtomicSense is that, instead of looking at individual objects, AtomicSense looks at the whole picture to identify it.

"We look for electromagnetic signatures, signals that are coming back - all at the same time," says Rezvani.

"You want to be able to identify a pedestrian walking with a dog in the middle of the street up to 300 meters away. You want to do that because when you are driving at forty-to-fifty miles an hour on a street, you do not want to have an accident because, by the time other sensors pick that up, it is way too late," he notes,

He calls previous radar systems Radar 2.0 and NPS AtomicSense Radar 3.0.

"Atomic neural processing is what we do. In terms of noise performance, it is fundamentally orders of magnitude better than what is currently state-of-the-art," says Rezvani.

NPS created a Level Four Autonomous sensor fusion platform to test multiband wide aperture radar along with cameras and LiDAR using its AtomicSense platform.

AtomicSense Radar vs LiDAR

"AtomicSense radar detection and detection is ten times better than traditional radar and doing as good a job as LiDAR," says Rezvani.

"Our focus is primarily on radar because we think that cameras will get better and better. So over time, there might not be a need for both camera and LIDAR in the equation. But that could potentially change in the near future. You cannot replace radar with any other sensors because of the way radar works - it solves many problems that neither LIDAR nor cameras can," he adds.

The AtomicSense platform is a patented software platform based on the Atomic Norm. It can run on a low-end GPU, a central processor or through edge computing. He says LiDAR is expensive because it is based on mechanical optical pieces that have to be very accurate. LiDAR is sensitive to dust and expensive to maintain.

Another advantage of the NPS AtomicSense radar processing is that it has fewer false positives than traditional platforms and provides more data points. He showed a company slide that states NPS AtomicSense sees farther, clearer and faster. It is safer in more situations with higher reliability.

"We can detect cars going across an intersection at 600 meters," says Rezvani. The radar detection can be 2.5x the distance of traditional radar.

The Future of NPS AtomicSense

NPS continually tests its sensor platform.

"We use the platform to bring all the sensors together and continue to work on where there is interest. We have a multi-platform system that we can look carefully at all the scenes and decide which one needs improvement and there is interest and demand by our customers that they want to see radar," says Rezvani.

"Our biggest value-add to the whole ecosystem of autonomy is on the radar side. Radar is the weak link right now. We are hitting the ball out of the ballpark with our software." 

NPS can help either for defence applications such as Raytheon or automotive with the software on signal processing, he reports.

With AtomicSense, customers can probably use fewer radars to achieve the same performance or potentially replace the LIDAR for cost savings, says Rezvani.

"Our ultimate goal is to save lives. We have already talked with several customers. It could AD-specific customers, OEMs or Tier Ones. Also, some of the defence companies because they are also interested to see things faster, clearer, better," he concludes.