G.fast is a digital subscriber line (DSL) standard for local loops shorter than 500 m, with performance targets between 150 Mbit/s and 1 Gbit/s, depending on loop length. High speeds are only achieved over very short loops. Although G.fast was initially designed for loops shorter than 250 meters, Sckipio in early 2015 demonstrated G.fast delivering speeds over 100 megabits nearly 500 meters and the EU announced a research project.
Formal specifications have been drafted as ITU-T G.9700 and G.9701, with approval of G.9700 granted in April 2014 and approval of G.9701 granted on December 5th 2014. Development was coordinated with the Broadband Forum's FTTdp (fiber to the distribution point) project. The name G.fast is a recursive acronym for fast access to subscriber terminals; the letter G stands for the ITU-T G series of recommendations. Limited demonstration hardware was demonstrated in mid-2013. The first chipsets were introduced in October 2014, with commercial hardware expected in 2015, and first deployments planned for 2016.
Frequency spectrum of the G.fast standard compared to VDSL2
In G.fast, data is modulated using discrete multi-tone (DMT) modulation, as in VDSL2 and most ADSL variants. G.fast modulates up to 12 bit per DMT frequency carrier, reduced from 15 in VDSL2 for complexity reasons.
The first version of G.fast will specify 106 MHz profiles, with 212 MHz profiles planned for future amendments, compared to 8.5, 17.664, or 30 MHz profiles in VDSL2. This spectrum overlaps with the FM broadcast band between 87.5 and 108 MHz, as well as various military and government radio services. To limit interference to those radio services, the ITU-T G.9700 recommendation, also called G.fast-psd, specifies a set of tools to shape the power spectral density of the transmit signal; G.9701, codenamed G.fast-phy, is the G.fast physical layer specification. To enable co-existence with ADSL2 and the various VDSL2 profiles, the start frequency can be set to 2.2, 8.5, 17.664, or 30 MHz, respectively.
G.fast uses time-division duplexing (TDD), as opposed to ADSL2 and VDSL2, which use frequency-division duplexing. Support for symmetry ratios between 90/10 and 50/50 is mandatory, 50/50 to 10/90 is optional. The discontinuous nature of TDD can be exploited to support low-power states, in which the transmitter and receiver remain disabled for longer intervals than would be required for alternating upstream and downstream operation. This optional discontinuous operation allows a trade-off between throughput and power consumption.
The forward error correction (FEC) scheme using trellis coding and Reed-Solomon coding is similar to that of VDSL2. FEC does not provide good protection against impulse noise. To that end, the impulse noise protection (INP) data unit retransmission scheme specified for ADSL2, ADSL2+, and VDSL2 in G.998.4 is also present in G.fast. To respond to abrupt changes in channel or noise conditions, fast rate adaptation (FRA) enables rapid (<1 ms) reconfiguration of the data rate.
Performance in G.fast systems is limited to a large extent by crosstalk between multiple wire pairs in a single cable. Self-FEXT (far-end crosstalk) cancellation, also called vectoring, is mandatory in G.fast. Vectoring technology for VDSL2 was previously specified by the ITU-T in G.993.5, also called G.vector. The first version of G.fast will support an improved version of the linear precoding scheme found in G.vector, with non-linear precoding planned for a future amendment. Testing by Huawei shows that non-linear precoding algorithms can provide an approximate data rate gain of 25% compared to linear precoding, however the increased complexity leads to implementation difficulties, high power consumption, and more costs.
In tests performed in July 2013 by Alcatel-Lucent and Telekom Austria using prototype equipment, aggregate (sum of uplink and downlink) data rates of 1.1 Gbit/s were achieved at a distance of 70 m and 800 Mbit/s at a distance of 100 m, in laboratory conditions with a single line. On older, unshielded cable, aggregate data rates of 500 Mbit/s were achieved at 100 m.
Service rate performance targets over 0.5 mm straight loops
|<100 m, FTTB||500-1000 Mbit/s|
|100 m||500 Mbit/s|
|200 m||200 Mbit/s|
|250 m||150 Mbit/s|
A straight loop is a subscriber line (local loop) without bridge taps. The listed values are aggregate (sum of uplink and download) data rates.
The Broadband Forum is investigating architectural aspects of G.fast and has, as of May 2014, identified 23 use cases. Deployment scenarios involving G.fast bring fiber closer to the customer than traditional VDSL2 FTTN (fiber to the node), but not quite to the customer premises as in FTTH (fiber to the home). The term FTTdp (fiber to the distribution point) is commonly associated with G.fast, similar to how FTTN is associated with VDSL2. In FTTdp deployments, a limited number of subscribers at a distance of up to 200–300 m are attached to one fiber node, which acts as DSL access multiplexer (DSLAM). As a comparison, in ADSL2 deployments the DSLAM may be located in a central office (CO) at a distance of up to 5 km from the subscriber, while in some VDSL2 deployments the DSLAM is located in a street cabinet and serves hundreds of subscribers at distances up to 1 km. VDSL2 is also widely used in fiber to the basement.
A G.fast cabinet on top of the copper cables cabinet.
A G.fast FTTdp fiber node has the approximate size of a large shoebox and can be mounted on a pole or underground. In a FTTB (fiber to the basement) deployment, the fiber node is in the basement of a multi-dwelling unit (MDU) and G.fast is used on the in-building telephone cabling. In a fiber to the front yard scenario, each fiber node serves a single home. The fiber node may be reverse-powered by the subscriber modem. For the backhaul of the FTTdp fiber node, the Broadband Forum's FTTdp architecture provides GPON, XG-PON1, EPON, 10G-EPON, point-to-point fiber Ethernet, and bonded VDSL2 as options.
Former FCC chief of staff Blair Levin has expressed skepticism that US ISPs have enough incentives to adopt G.fast technology.