Massive MIMO explained in depth

Eom Marzetta invented Massive MIMO in 2010 and the first production units shipped in 2017 in Japan & China. As the build of 5G mid-band picked up in 2019 and 2020, M MIMO became the standard choice. As I write in late 2022, China has deployed over two million cells.

MIMO uses multiple antennas to send several signals in the same spectrum. With 4 antennas, typical speeds double. Results vary based mostly on the terrain.

M-MIMO was the next step. The typical M-MIMO uses 32-64 relatively small antennas. Lower frequencies (below ~2300 MHz) require larger antennas so it is generally impractical to have more that 4 or 8 antennas so rarely implement M-MIMO.

Continue reading

Primary: Cost per bit of wireless is falling at a ferocious rate

Even experts are skeptical that Verizon’s cost per bit is falling at 40-50% per year, but I’ve confirmed it to my satisfaction. Three well-established datapoints:

  • Traffic was growing at 60-80% per year and only now is growth falling below 40%
  • Capex has been flat to down
  • Speeds have been going up consistently and congestion is uncommon

If wireless cost had not been plummeting at a rate similar to traffic growth, without a rise in capex the network would seriously degrade. Ergo: Deployed technology is improving very rapidly.

Verizon CEO Vestberg calculates his cost per bit has fallen 90%. His peer at Deutsche Telecom estimates 70%. Neither figure is precise since neither offers details, but the trend is clear.

The trend has been clear at least since 2013, when carrier aggregation and MIMO became widely deployed. It will continue at least until ~2023, based on the equipment (Massive MIMO,) mid-band spectrum. and the likely maturity of SDN, NFV, O-RAN and the like.

At the same time, routers, switches, optical backhaul and other network components have been going down at a similar, extraordinary rate. Moore’s Law is slowing as chip features aproach atomic dimensions, but More than Moore techniques keep improving productivity.

All of which is wonderful, especially if carriers can find a way to sell what they can deliver.

Do also read Moffett’s Remarkable Insight: Low marginal cost means nothing if you can’t sell it for insight on the implications of technology advance.

Low cost can be an extraordinary weapon

Low costs, passed on to customers, can re-create the market. The biggest telecom story of the last decade is the 400 million Indians newly with 4G. Mukash Ambani spent US$30 billion for a state-of-the-art, extraordinarily efficient network covering 95% of 1.353,000,000 Indians.

With prices from $2-$10, he rapidly won customers from the 10 other Indian telcos. Most were selling 2G and 3G, saying 4G was too expensive for India. This was hogwash; 4G, the more recent technology, is an order of magnitude less expensive. Ambani makes money at prices so low all but two other companies are gone. Survivor Bharti wants a government handout and Vodafone is on the edge of bankruptcy.

Once the network was built, Ambani’s cost per bit or additional user was very low. He took that to the extreme, literally giving away three months or more of service to new customers, without catches. Ten million signed up each month and soon became paying customers. Jio is now at scale, profitable, and soon the most comprehensive “platform” on earth.

Ambani is extremely proud he is creating the first great Indian Internet company.

Low costs are not a guarantee of success

The economics textbooks suggest being the low-cost provider is a crucial strategy. This isn’t necessarily true in telecom, where fixed costs are high and marginal cost only a modest factor. See Moffett’s Remarkable Insight: Low marginal cost means nothing if you can’t sell it

Rakuten in Japan, Dish in the US, 1&1 Drillisch in Germany, and possibly Oi in Brazil are hoping to replicate the success of Jio. The other companies in India were just getting started on 4G. Most were selling 2G & 3G, and most consumers thought 4G phones too expensive. But 4G smartphones had fallen to US$50. Ambani asked for bids on tens of millions of 4G feature phones. That brought the phone price down to ~$25. It makes sense to find ways to get all customers on 4G, because they will cost much less to service.

In Germany, Japan, the U.S. and Brazil, the empires can strike back powerfully

The new 4th carriers will have a more modern network than even giants like Deutsche Telekom and NTT DOCOMO. But the incumbents are in far better shape than the 2G and 3G networks Jio wiped out. Most have 90-98% 4G coverage and are starting to go beyond pr in 5G.

Even in Brazil, customers of TIM get a 4G signal 88% of the time. Claro and Vivo are not that far behind. Vivo has passed 13 million homes with fiber, which also provides robust wireless backhaul. If Marc Ganzi takes over Oi, Brazil’s $4, he will face strong competitors.

If the newcomer is successful, the incumbents can and will strike back. Verizon and AT&T typically make $20 billion a year. NTT DOCOMO, KDDI,DT, Telefonica; and American Movil/Telemex make billions. Their capabilities are immense.

If Verizon raised capex by 3% of its $125 billion in sales for 3 years, that would add $10 billion, enough to rebuild most of the network to latest and greatest. Actually, they’ve been doing that already. Verizon’s “One Fiber” program cut network costs by 50% in the first year alone, Lee Hicks told me.

Even with advanced technology, being the fourth carrier is tough. The main opportunity is when the first three carriers have cartel-like pricing, which is common.

Virtualization savings are not enough

Rakuten’s Tareq Amin believes his new Japanese network is costing 30% less. It’s cloud-native, software defined, fully virtual, and uses a software RAN. It’s got a way to go, but let’s start with his 30% figure How much difference does that make?

Let’s assume the marginal cost of serving a new customer is about 15% of the revenue, once customer acquisition costs are covered. A saving of 30% would lower the cost from 15% of revenue to 10% of revenue. That’s about 5%. It’s meaningful but usually dwarfed by marketing and customer acquisition costs.

The engineers actually building networks , like Alex Choi, are enthusiastic about the potetial but are reporting numerous challanges still to solve.

The big productivity improvements came from MIMO, Carrier Aggregation, Massive MIMO, rebuilding the network inside the carrier, and more spectrum

The huge productivity improvements began around 2013 in 4G and will continue driving productivity at a frantic pace at least until 2022-2023.

There’s a myth that 5G NR yields major capacity improvements. The highest plausible estimates come from T-Mobile CTO Neville Ray, He swore under oath at the FCC that 5G NR could result in 19% to 52% improvement over what 4G can deliver today using the same spectrum. 4G LTE is so close to Shannon Law limits it’s impossible for any software to massively increase capacity.

Anyone claiming otherwise is ignoring the progress made in 4G. Between 2013 and 2020 of the typical 4G network went up 10-25 times while capex was flat to down. If demand was greater, most networks could have done twice as much while still keeping capex reasonable.

MIMO

Paulraj’s invention produced about a 2-4x improvement starting around 2012 by using 4-8 antennas. The big deployments were from ~2017. Neither DT nor AT&T has deployed MIMO everywhere and room for improvement remains. The results vary by terrain.
Typically, a 2-4X improvement.

Carrier Aggregation

LTE was designed to use 20 MHz carriers. By around 2013, LTE Advanced allowed joining carriers, initially 2 or 3 but now up to 6. A company like AT&T might combine spectrum at 700, 1799, 1900, and 2300 MHz. Five bands at 3.5 GHz would be possible in 4G, although most of 3.5 GHz from telcos will be 5G.
AT&T is actively combining Wi-Fi/unlicensed spectrum at 4.9 GHz, sometimes doubling performance. This works in 4G but not yet in 5G, one reason 4G is often faster than the 5G deploying today.
Typically, a 3-5X improvement is possible.

Massive MIMO

Tom Marzetta invented M-MIMO in 2010 at Bell Labs, but Huawei and ZTE were the first to deploy in 2016. The results from Softbank Japan and China Mobile in 2016 were improvements of 3X-10X in TD spectrum, widely varying by terrain. Nokia and Ericsson couldn’t ship M-MIMO in quantity until 2019, so some people think it is 5G. Actually, 5G offers only very small improvements over 4G.
Massive MIMO is key to opening up 900 MHz of spectrum, almost doubling the available spectrum. Without Massive MIMO, the throughput for mobile data in mid-band was not enough to cover the cost. Now, that spectrum is selling for billions.
M-MIMO works poorly in FD bands, below 2300 MHz. Some hope for a technological breakthrough.
Typically, M-MIMO results in a 3-5X improvement in TD spectrum.

Other technology improvements (Dozens)

Coding more bits per Hz (256 QAM) raised capacity by almost a third starting in 2016. It’s still not rolled out everywhere. SON – self-organizing networks – reduce interference between cells. Ibrahim Gedeon of Telus told me in 2018 that made an enormous difference.

More spectrum

Opening the band from 3300 MHz to 4200 MHz is a big advance, doubling the available spectrum. Even more important was the effect of carrier aggregation on the use of bands telcos already owned. Carriers like AT&T have been warehousing spectrum for a decade, mostly to prevent competitors from buying it. (Ask any expert on spectrum auctions.)
As recently as 2018, AT&T’s CFO told Wall Street half of its spectrum was fallow. (No one believed that but I confirmed similar with several carriers.) In fact, AT&T was actively soliciting buyers for its spectrum at 2300 MHz because without carrier aggregation it was impractical to use. As 3-5 carrier aggregation became practical, and phones began to support it, the effective spectrum at many carriers increased 50% or more.

In 2014, world-class engineers debated whether wireless would soon see a 10X or 100X capacity improvement

In a remarkable symposium accompanying Paulraj’s Marconi Award, Paul, John Cioffi, and others extrapolated wireless progress. Vint Cerf, Tom Marzetta, and a dozen other luminaries took part. For the last six years, I’ve watched the vendors deliver much of what the pioneering engineers predicted.

At the dinner for the award, Paulrj quietly told me, “I’m confident about 50X, but not so sure about 100X.” I bet if he looked forward again, he’d expect the 100X this decade.