Verizon and Nokia are ready to test. Nokia CTO Hossein Moiin intends to test Antonio Forenza's pCell technology early next year. Last year, I wrote extremely skeptically about Artemis/pCell informed by three very respected engineers. CEO Steve Perlman made wildly implausible claims, including that they would deploy across 350 San Francisco rooftops by the end of last year. They haven't been seen.

Moiin is a respected, independent engineer. I have to look again. USC Prof Giuseppe Caire now says "The early trials showed pCell achieving far higher concurrent user capacity than any wireless technology I am aware of." That's very different from his earlier comments (below) and I assume he's been shown something newer. Peter White at Faultline has an important article about semi-secret trials in U.S. sports stadiums. (Paywall.) 

Verizon in a startling move Monday told analysts they would soon test Massive MIMO and beamforming. (pCell is very similar.) These are key "5G" tools that weren't expected until about 2020. 

The Verizon interest may explain why Nokia was so supportive of pCell in the press release. 

Very big improvements are close in wireless. Arogyaswami Paulraj, Henry Samueli, Andrea Goldsmith and Vint Cerf in Marconi seminars were comfortable with the forecast wireless capacity would soon increase 50 times or more. Massive MIMO will play an important role.

Nokia will test in a stadium, where most receivers will be stationary. Line of sight is excellent. Power, cooling and space can be found. That's a far easier situation than the unrealized plan to go to rooftops across San Francisco, although still not trivial. 

CTO Antonio Forenza's engineering work is in the mainstream of wireless research, MU/Massive MIMO and interference cancellation. Forenza's previous experience and a brief conversation convinced me his engineering background is strong. He worked (briefly) for Paulraj at Iospan and Marty Cooper at Arraycomm, both wireless engineers who have won the Marconi Prize. He went on to a doctorate under Robert Heath at the University of Texas - in just this kind of wireless engineering. Heath is a leader in MU MIMO and has a testbed in Texas.

Artemis' connection of this kind of MIMO to existing LTE phones was impressive and as far as I know unprecedented. Now, pCell has developed a 35 antenna hub, just about right for beamforming to the 16 mobiles in their improved demo. (16 phones in picture which were connected to a 35 antenna hub.) In a small space, they ran each of them at 12.5 megabits for about 200 megabits in 5 MHz. 8x8 MIMO could just about do that and is in the LTE standard. But I don't think 8x8 is in the field, so if this proves practical it's impressive. An 8x result is excellent; I assume the 35x in their release is not an appropriate comparison but I haven't had a chance to doublecheck that with a good engineer.

Perlman has a reality distortion field second only to his mentor, Steve Jobs. He convinced numerous journalists (even the NY Times) he would very quickly transform wireless. Once a century a Ramanujan comes along and transforms a field. That's about what it would take for everything Perlman suggested. The performance of his "amazing" Columbia demo could have been delivered with existing LTE MIMO and comparable had been. It was as convincing as what the Wizard of Oz had claimed -- before he was found behind the curtain.

Going into testing in 2016 for a limited application is far more realistic than a production system in 2014. Tom Marzetta of Bell Labs has done important work in this area and believes there's still a way to go before it's ready. pCell is suggesting they have made important progress in active interference cancellation, a technology that may be close. It's been in the textbooks for years but rarely implemented on this scale. The calculations were beyond what was practical. 

Maybe Moore's Law will change that. This demo required three servers for 16 receivers, too much for most applications. Moore's Law - or custom chips - will reduce that burden. Henry Samueli at Broadcom mentioned in a seminar they were working on chips to drive 50 or more antennas, but they haven't released information beyond that. Tower Semi actually has a 256 antenna chip, developed with UCSD. That one is probably aimed at expensive DOD applications.

John Cioffi faced a similar problem when he invented vectoring, noise cancellation for DSL. In 2003 he and George Ginis wrote the first paper on vectoring, which they knew was impractical at the time. John estimated Moore's Law would lead to processors with the needed capacity around 2010 if I remember correctly. Volume shipments for vectoring didn't begin until 2014 but at least ten million lines are on the way.

If server capacity has held things like this back in wireless, now is the time to jump on and trust Moore's Law. Henry Samueli is a prominent skeptic on Moore's Law as it reaches atomic scale, but even he believes we have chip generations still to come.   

This patent below is almost certainly not what Forenza is using. I include it to point to the long running and sophisticated work that has been going on on interference for over a decade. Forenza may have significant improvements but I lost confidence when CEO Perlman made implausible claims. In particular, I heard him claim revolutionary unprecedented inventions in areas a Stanford Professor had previously described to me work going back years. The Times picked up an error about the demonstration that came from Perlman.    

Nokia Networks, Artemis to trial pCell technology

• Companies sign MoU to collaborate in the prototyping and trialling of Artemis pCell wireless technology
• Artemis Networks LLC is a wholly-owned subsidiary of Rearden LLC

Espoo, Finland - 2 November 2015

Nokia Networks and Rearden LLC have signed a memorandum of understanding (MoU) to jointly test Artemis™ pCell™ wireless technology in 2016 with wireless operators, initially in large indoor venues and other high density areas. pCell has the potential to enhance the capacity of conventional 4G TD-LTE networks in certain use cases, while remaining compatible with unmodified devices.

Under the agreement, the companies will jointly offer pCell Proof-of-Concept deployments to selected Nokia Networks customers. The companies may extend the collaboration to consider further advanced features that could be enabled by pCells, such as precise 3D location positioning.

pCell wireless technology highlights:

• pCell technology brings a radical new approach to wireless that is intended to exploit - rather than avoid - interference to synthesize a tiny personal cell (a “pCell”) for each wireless device.
• This enables each device to use the full capacity of spectrum concurrently, rather than taking turns sharing spectrum with other devices.

Hossein Moiin, Executive Vice President and CTO at Nokia Networks, said: “In addition to creating transformative internal innovations, we continuously look for and evaluate external innovations to bring the most advanced solutions to the operators. We are keen to see the potential for pCell in enhancing 4G LTE downlink and uplink capacity given the rapidly growing network demands such as concurrent HD video streaming.”

Steve Perlman, Founder & CEO of Artemis Networks, said: “We are delighted to collaborate with Nokia, the world’s leading 4G LTE network vendor, to offer pCell Proof-of-Concept deployments to selected customers. pCell technology has the potential to significantly increase the downlink and uplink capacity of spectrum, while remaining compatible with existing 4G LTE devices.”

Giuseppe Caire, Advisory Board Member of Artemis Networks and Professor at USC Viterbi School of Engineering and Alexander von Humboldt, Professor at the Technical University of Berlin, said: “The early trials showed pCell achieving far higher concurrent user capacity than any wireless technology I am aware of.”

For your information, DIDO is *nothing new* (a part from the funny acronym) and has been widely investigated for at least 10 years by a large number of researchers. This is just a special case of the so-called vector broadcast channel, for which the Shannon capacity has been determined (including some preliminary award-winning work of myself) and it scales as M * log(SNR) for large SNR, where M is the minimum number of transmit and receive antennas, even though joint processing is done only at one side (the transmitted side, via the central server coordinating all the access point antennas). Therefore, comparing the rates achieved by the vecor broadcast channel with the simple log(1 + SNR), which is the Shannon capacity formula for a single point to point Gaussian channel, makes no sense. Of course, joint processing of the transmit antennas in ``DIDO'' (let's call it this way ...) does not beat Shannon! A theorem proves that this cannot be done. Only, you have to use the correct capacity formula for this scenario, which is *not* log(1 + SNR). The DIDO guys just ``forgot'' to put the multiplexing gain M in front of the log SNR term!!!

Any graduate student working in information theory and communication theory would immediately realize this ridiculous mistake. Claiming novelty for the idea is completely non-sense, since as I said this stuff is well-known and has been around for 10 years in the scientific literature and also in various industrial prototypes. 

What is new, and worthwhile attention, is the implemnetation: if the Rearden guys managed to perform joint processing of a large (meaning 10 to 50) number of access point antennas through a DSL link, solving the problem of phase and timing asynchronism of the separated terminals, and **without clocking everything through a lab-grade common clock, then they do have a point. 

By the way, in my Lab at USC we do have such a prototype, for a small number of transmit antennas (today we can do this for 4 antennas on 4 separately clocked radio cards). 

In conclusions, please stop this non-sense and let's start asking the DIDO guys the relevant questions: can you do this in a scalable way through DSL links, in order to serve people's homes?
do you require a common precise clock distribution, or can you synchronize in some other way, with commercial grade low-cost components? What type of downlink precoding do you actually implement? How did you solve the problem of the finite-rate backhaul network, e.g., DSL links, without placing an oversized fiber optical channel connecting the terminals to the central server?

These are the questions that need to be answered to make distributed multiuser MIMO (this is the name under which ``DIDO'' is known in the scientific literature) a promising technology in the real world.