Electrical power control system repeatedly knocks out G.fast. While vectoring can reduce most interference, it does not eliminate all of it. TeliaSonera was disappointed as "user links were frequently interrupted, severely affecting G.fast performance."
Using a signal analyzer, they found the cause was an instantaneous high-amplitude pulse interference from an electrical power control system located near the basement. Fortunately, the standard allows multiple mode determined controlled by software. They were able to solve the problem with anti-pulse interference processing.
Telcos remain hopeful that customer self-install will be fine. Given that few customers own a high-speed signal analyzer, it's not clear how a case like this would be handled. Those in charge of field trials are closely watching how much speed is lost when the customers self-install. The loss is meaningful on average and in cases like this highly significant.
John Cioffi of ASSIA has been pointing out that a decent average performance is not good enough if some customers are badly affected. He believes leaving 5% or 10% of homes with problems is unacceptable. So do the people in those homes. In the USA, telcos have generally decided that delivering terrible service to a small percent of customers is cheaper than fixing the problem. It's time for performance measurement to include a 95% test, not just an average.
The G.fast signal was cut off at 23 MHz to (successfully) allow ADSL and VDSL (17 MHz) in the same binder. The top end was cut at 88 MHz to avoid interference from FM radio. Speed dropped in half.
Trevor Linney at BT is running the largest test of G.fast and everyone is hoping to hear his results.
The Telia details are from a detailed technical discussion by Huawei.