Use the plastic sheathing around the phone wires as a waveguide. With 1,000x more spectrum, you get 1,000 more speed. Glass fiber optic waveguides can carry 250 terabits. John Cioffi wants to apply similar techniques using the air gaps between the plastic surrounding a billion phone lines.
The proposed terabit DSL can use 300 GHz+ of spectrum, "submillimeter waves." Current DSLs use 100-500 MHz. Higher frequencies just wouldn't make it through a standard copper wire. The signals get weaker (attenuate) very quickly in copper. 500 MHz can only carry about 30 meters. Gigahertz, even less.
Cioffi proposes using the tiny air spaces between the plastic insulated wires as a "waveguide." The signal would travel over the air gaps, not the copper or plastic. Fiber optic, glass or plastic, "guides" the waves; why couldn't the plastic insulation do similar? (The waves are very, very small. They can easily fit in the gaps.) The signal is carried through the air between the wires, not on the copper wires.
Simple explanation of terabit TDSL: Very high frequencies - 500 GHz to a terahertz or more - can carry enormous amounts of data, up to a terabit over short distances. The plastic insulation around existing phone lines could "guide" those signals. That's similar in some ways to the role played by fiber optics. The DSL highest speeds demonstrated over copper are 11 gigabits over very short range. TDSL would run in the gaps between copper pairs, not over the copper.
There's major research to do even to build a demonstration model, much less practical systems. Curves and irregularities in the cable binder need to be overcome. Vectored DSL and 5G wireless beamforming have proven that much can be achieved with multiple antennas and very high-speed calculations, but this goes beyond anything previously demonstrated.
5G millimeter wave wireless runs at 26 GHz (Europe,) 28 GHz (U.S.) and 39 GHz. WiGIG is 60 GHz and commercial microwave often 70 GHz - 90 GHz. 300 GHz and 500 GHz is only used in a number of research labs and perhaps military applications. It's not available commercially, one reason building a demonstration unit will be a major task.
Multiple miniature antennas will direct the beam, in a way similar to massive MIMO. They will require massive processing power. One researcher believes the latest generation of GPUs might be sufficient. Others assume another iteration of Moore's Law will be necessary. Fortunately, Moore's Law has at least two more steps. Many, including Marconi Prize winner Henry Samueli, believe the future of Moore's Law is cloudy beyond the 5 nanometer generation.The chips will be approaching the size of single atoms.
Cioffi's presentation is at http://www.assia-inc.com/wp-content/uploads/2017/05/TDSL-presentation.pdf. Co-authors are Ken Kerpez, Chan Soo Hwang, and Ioannis Kanellakopoulos. He's shared it with numerous respected engineers. They tell me they are impressed, although nothing is certain until the hardware for testing is built. Several dozen engineers have reviewed the work, identifying some practical problems that would need to be solved. No one, as far as I know, found fundamental flaws in the theory.
In 1993, the best engineers in the field believed DSL modems could not go beyond 1.5 megabits. John Cioffi invented DMT line coding and delivered a modem running at 6 megabits down. Two hundred million homes are now connected with DMT ADSL. A few years later, John again went beyond the state of the art, using Dynamic Spectrum Management to deliver more reliable speeds over DSL. In 2002, he introduced "vectoring," essentially noise canceling for DSL. The chips of the day couldn't keep up with the necessary calculations. John hoped that chips would catch up by 2010; in fact, it took until 2013. Since then millions of lines have shipped and tens of millions are promised.
That's three near-miracles Cioffi has delivered. He's hoping for a fourth.