AT&T is so big people in different parts of the company often have different opinions. An insider told me T was going to use millimeter wave widely, believing it would attract enough customers to pay off. He saw it as a land grab situation against Verizon, who is firmly committed to tens of millions of homes passed. "Whoever is first with 5G millimeter wave will win 10% to 20% of customers away from cable. We will be in a race to be first in many parts of the country." Their technical people have been preparing a large build. At least one thought the company strategy would be like Verizon, deploying quickly to get that first mover advantage.

The man with the money is dubious. He's more inclined to grow fiber home where they already have fiber nearby. It's "Very inexpensive." He wants to focus on building fiber to the 30M homes passed by U-Verse in district. They've already fibered 10M of them and are doing 3M/year. Broadband in district has been flat to down. They are getting clobbered where they haven't upgraded. So protecting a "fortress AT&T" is a sensible strategy.

Karri Kuoppamaki tells me T-Mobile has major plans for Massive MIMO in frequencies below 2300 MHz, which are designated for FDD (Frequency Division Duplexing.) TDD Massive MIMO has been shipping by the tens of thousands since 2016, but FDD is still only in trials.

Massive MIMO was developed for TDD, which sends both upstream and downstream in the same frequencies. FDD uses separate frequencies, which makes it harder for the phone to provide the base station information it needs for best performance. 

Professor Erik Larssen believes the signaling problem is so profound FDD may never be effective. But at least 8 telcos are in FDD trials. None has moved to deployment. 

TDD performance is estimated at 3X (China Mobile,) 5-7X (India,) and 
up to 10X (Softbank Japan.) Terrain has an enormous effect, so speeds are very different within and across networks.

T-Mobile is seeing somewhat lower rates, as expected.

TSMC is rushing to produce 10's of millions of chips for the next iPhone, taking up nearly all their 2018 capacity at the new 7 nm node. Qualcomm and possibly Huawei have the chips at an advanced stage of design and should have sample chips in the next few months.

Rick Merritt at EE Times reports, "TSMC is in volume production of 7-nm chips today with more than 50 tapeouts expected this year. It’s making CPUs, GPUs, AI accelerators, cryptocurrency mining ASICs, networking, gaming, 5G, and automotive chips."

.

Timotheus Höttges & John Legere bet their merger on bamboozling DC on "What is 5G?" They are using New Definition 5G, introduced in 2017 and 70% to 90% slower than the good stuff, millimeter wave.

Legare is swearing to Washington the T-Mobile 640 MHz 5G NR network will be many times faster than LTE. That isn't true, of course. It's far more likely to be only 25%-50% faster, or perhaps less. It may even be slower than the 500 MHz LTE/LAA T-Mobile already has in Manhattan.

In 2016, 5G meant millimeter wave and (for some) Massive MIMO*. Millimeter wave bands had far more spectrum available, 400 or 800 MHz channels compared to 100 MHz maximum in LTE. 20 gig peaks work in the laboratory and Verizon is delivering a true gig to customers in 2018.

Folks building LTE-speed networks wanted to be called "5G" and take advantage of the massive hype. So they made "New Definition 5G" with a pr campaign and a minor software tweak, NR (New Radio.) 4G LTE networks became "5G." Every engineer in the business knows this is a scam.

T-Mobile in the U.S., China Telecom, and many others are deploying 4G speed networks and calling them 5G. They are 70% to 90% slower than the originals.

T-Mobile's "5G" network will be built in 600 MHz spectrum and will only be 25-50% faster than its 4G LTE network, perhaps 400-700 megabits. Ted Rappaport of NYU Wireless, a world-class expert on 5G, emails me:

The low band (600 MHz) systems will always be limited in capacity and are not future proof no matter what special massive Mimo is used- the antennas will be too large and the bandwidths are too small, and simply can't carry the bandwidths of mmwave-- the RF channel allocations are simply too small. The wider bandwidth channels at mmwave is the only way to carry the multi Gbps data rates that will be seen in 5G. And to simultaneously also provide backhaul between cells in the same network.  

The secret sauce in millimeter wave is 400 MHz and 800 MHz channels. The 600 MHz and other low band use 20 and 40 MHz. Tmo' 600 MHz will probably be 400-700 megabits. mmWave is delivering 10gig and 20 gig in the laboratory. Verizon is getting 1 gig to each customer in their trials and can go higher. 

In 4G LTE/LAA, Tmo's 4G is delivering 500 megabits to live customers in Manhattan, pretty darn good. To do much better than that in 20 MHz 5G NR would require breaking the Laws of Physics.

Craig Moffett recently wrote, "The probability of regulatory approval at 50%, the odds of deal consummation are therefore something like 40%." At Bernstein, Moffett was perennially rated the #1 analyst on Wall Street when at Bernstein. He's remarkably creative. He's particularly well informed about D.C. Since at least 2009, he's made a point of actively building D.C. contacts at the highest level. FCC Commissioners care about Wall Street opinion and have often looked to Craig.

Jonathan Atkin of RBC agrees. "We continue to believe the likelihood of a Sprint/T-Mobile deal approval is less than 50%." Walter Piecyk believes, "There is less than 40% chance the Sprint/T-Mobile deal is approved by regulators."

My take: If D.C. gets the facts, the chance is less than 10%. If government people drink the Kool-Aid, it might just go through. 

Karri Kuoppamaki, VP Technology at T-Mobile, startled me with the answer, “My vendors tell me NR will be 25% to 50% faster than LTE.” A second source, with similar information, expects about 20%. Neither have much actual test data, and vendor salesmen are often over-optimistic. But even 20% is more than my previous reporting and I need a correction if that proves out.

Eric Xu, current Huawei Chairman, recently said consumers would find no “material difference between 4G & 5G.” That remains true. Peak speeds will be similar and total capacity only matters if the cell is near to congestion, surprisingly rare. Even then, is a fall from 300 megabits to 200 megabits meaningful to the consumer?

A 25% improvement is not a new generation.

26M added on the most efficient network in the world. Open Signal testers were able to access an LTE signal 96.4% of the time. That's more than comparable testing on any of the U.S. carriers. Jio pledges to reach 99% of the 1.3B Indians by the end of 2018.

The secret sauce is a state of the art all-IP 4G network that delivers data for ~$0.05 per marginal gigabit. The network cost > $20B to build, but once in place has very low marginal costs. The 150,000 base stations are two or three times as many as Verizon or AT&T, many with Massive MIMO.

They control 300,000 kilometers of domestic fiber and a share of 25,000 kilometers of undersea international.