MilliLabs Ignores Industry Skepticism to Build Emulator for Millimeter Waves

The startup says it has an alternative to expensive over-the-air tests for 5G

4 min read

Back of emulator showing wires
Photo: MilliLabs

Throughout its development, 5G has been plagued by a simple problem: Is there a way for engineers to test millimeter wave propagation without committing to expensive and complex methods? The founders of one startup, MilliLabs, say they’ve found a solution.

“Everyone was doing over-the-air testing and we’re saying, ‘Dude, that’s nuts,’” says Aditya Dhananjay, the co-founder and president of MilliLabs. It was widely assumed that emulating millimeter waves was, for all intents and purposes, impossible.

Prior to 5G development, standard industry procedure was to use channel emulators to quickly gather a large amount of general data on the technologies being developed, before conducting more refined, over-the-air tests.

Traditionally, emulators provide a way to test how signals will propagate through an environment without needing to actually go out and transmit any signals. That means they use cables in place of antennas, send the test signal through a digital emulation—or channel model—of the environment in question, and receive the results through another set of cables in place of the receiving antennas.

That’s simple enough for lower-frequency, far-traveling radio waves, which only need one, two, or maybe four antennas to both transmit and receive. But 5G will employ higher frequencies—millimeter waves—which require dozens, if not hundreds, of tiny antennas working together. That’s because millimeter waves have the annoying tendency to disperse quickly and get blocked by people and cars, not to mention the sides of buildings.

Millimeter wave signals will rely on beamforming to be useful, which tightly focuses and strengthens a signal in a desired direction. Dhananjay compares it to ripples in a pond. If radio waves are like dropping a stone in a still pond to send ripples in every direction, beamforming millimeter waves will be like dropping dozens of stones in a precise pattern to cancel out ripples in some directions and increase the strength in others.

And lots of stones—or in actuality, antennas—would mean dozens, if not hundreds, of cables snaking into and out of the emulator. In other words, it’s complicated.

In December 2017, Maryam Rofougaran, one of the co-CEOs of Movandi, a wireless industry startup, said that over-the-air testing was “inevitable” as the standard method of testing millimeter wave signals. A whole host of problems—antenna design, signal construction, and more than anything, cost—make the building of test systems to emulate millimeter wave propagation in real environments vastly different from emulating lower-frequency radio waves.

“As of 2018, it is not so much about what the emulator can do but in what detail and accuracy.”

The most expensive 4G emulators, with four antennas in and four out, already cost half a million US dollars each. For 5G, with millimeter wave signals expected to require, at a conservative guess, 10 times as many antennas, the proposition becomes prohibitively complex and expensive.

Dhananjay says one back-of-the-envelope calculation for a relatively modest—for millimeter wave beamforming, at least—64 in, 64 out array produced an estimated cost 100 times higher than the most expensive 4G emulators. “It’s already a non-starter,” he says.

So MilliLabs decided to try emulating the antennas as well. The team reasoned that while it was possible to use dozens of analog cables to construct a beamformed signal, if you knew what the construction of the signal you wanted to test looked like, you could just define it mathematically and send that into the channel model.

Ultimately, their emulator sends only two pieces of information into the channel model—the mathematically defined signal and the direction it will travel through the emulated environment. Yet it can also emulate up to 1,024 antennas on each end, and handle bandwidth well over 2 GHz. For comparison, the biggest commercially available emulators have eight antennas on each side, and the best bandwidth that 4G emulators can do is 160 MHz.

MilliLab 5G millimeter wave emulationMilliLab’s emulator at work. After a signal has passed through an emulated environment, the four graphs give a sense of how clearly and strongly the signal was received.Photo: MilliLabs

Dhananjay says their solution drastically cuts down on the complexity of both the emulator’s construction and its cost. By only requiring a handful of cables, rather than the dozens needed to emulate each antenna individually, MilliLabs was able to construct its emulator using off-the-shelf parts from National Instruments, which sponsored much of their research.

“If the model describes a particular kind of channel, our emulator can emulate it.”

Wonbin Hong, a professor at Pohang University of Science and Technology in South Korea, says that any emulator that could do what MilliLabs claims theirs can do would drastically shorten the entire development cycle of millimeter wave technology. But he’s cautious.

“As of 2018, it is not so much about what the emulator can do but in what detail and accuracy,” Hong says. He explains that advances have been made to reasonably emulate small-scale or large-scale environments, like an office interior or a signal beamed between two base stations, but emulating mid-scale environments remains elusive.

“For instance, if the emulator cannot offer me a different response for a 5G emulation in downtown Chicago during the summer versus the winter or 3:00 PM and 2:00 AM, this is a problem,” Hong says. Even then, he believes the biggest hurdle for something like MilliLabs’ emulator will be to demonstrate that their results are consistent with the traditional, expensive, time-exhaustive method of over-the-air testing. “Nobody wants to make multi-million dollar business decisions based on unproven methods,” Hong says.

For his part, Dhananjay is confident in MilliLabs’ emulator. MilliLabs plans to sell customization services to partners who need specific channel models emulated. Dhananjay says that their emulator can handle the complex models thrown their way. “If the model describes a particular kind of channel, our emulator can emulate it.”

The Conversation (0)