G.fast technology

G.fast is a technology that communicates with a local fibre to a cabinet (FTTC) broadband point using signalling technology down a twisted pair to the premises then onwards to the internet by fibre. It does not need fibre running all the way to the premises or upheaval for installing it like a full fibre to the premises (FTTP) deployment.

Installation is managed by BT Openreach currently. The telephone socket is replaced with the latest NTE5C with MK4 prefiltered faceplate. The G.fast modem is connected to the line by BT Openreach. It is the Huawei G.fast MT992-10 for none BT broadband G.fast ISP’s. Sky will likely have their own G.fast router when they eventually resell ultrafast broadband?

G.fast estimate for our premises

G.fast VDSL2 comparision same twisted pair to cabinet

With VDSL2 I achieve 80Mbps/20Mbps synchronisation speed and 74.7Mbps/18.7Mbps throughput which was measured with a SamKnows Whitebox for superfast broadband on Sky Fibre Pro.

The G.fast modem communicates with the fibre cabinet using time division duplexing (TDD) compared to the technology in ADSL2+ & VDSL2 that uses frequency division duplexing (FDD). FDD is full-duplex transmitting and receiving concurrently whereas TDD is half-duplex either transmitting or receiving. 

With G.fast physical-layer retransmission and vectoring across the twisted cables in the bundle is mandatory this keeping the crosstalk, reflections and impulse noise impact minimal. It also uses seamless rate adaptation (SRA) for slow-moving synchronisation speed movements, which takes several seconds, and fast rate adaptation (FRA) to set the synchronisation speed dynamically across a group in <1ms on the fly; the latter primarily for interference reasons on the twisted pair. DLM is functioning on the line and monitors for a day and then takes action based on the previous day. The noise margin is 3dB for G.fast but the line must be performing perfectly to reach that in relation to retransmission perfectness.

It is important to note VDSL2, and G.fast latency with TDD does have differences for roundtrip delay (ping). FDD for VDSL2, the latter should be lower but for the fact, G.fast uses higher frequencies than VDSL2. Those who believe the roundtrip latency on G.fast is worse than VDSL2 because it is full duplex need to calculate the figures correctly? Roundtrip ADSL2+ for 180m is ~2.9ms peaking to ~15.6ms, roundtrip on VDSL2 is ~1.7ms peaking to ~15ms and ~0.1272ms peaking to ~1.536ms on G.fast broadband without interleaving for the same distance through DSLAM.

VDSL2 roundtrip latency to BBC from Plymouth 80Mbps/20Mbps synchronisation speed on VDSL2 to ECI cabinet about 15% bandwidth utilisation, distance 400km for fibre 4.9 milliseconds a km for 3.92ms delay roundtrip on fibre. 2ms by equipment on my LAN (3 routers and an ethernet bridge). 1ms for equipment on internet WAN. So, the total is 6.92ms from cabinet to London with equipment in the premises and the internet without the twisted pair.

Latency with VDSL2 from Plymouth to London

Actual latency on the twisted pair to the cabinet for 180 metres on VDSL2 is approximately 6ms on this telephone line to the cabinet. hopefully, those reading are keeping track with the calculations for real latency roundtrip to cabinet on a twisted pair for 180 metres long that is 6ms here mean!

So, I should see a drop of 4.544ms minimum on G.fast to the cabinet on the twisted pair even with TDD that doubles the roundtrip for 180 metres.

Even though VDSL2 is full duplex for achieving low roundtrip delay, G.fast signals will switch between transmit and receive quickly operating at frequencies between 19Mhz and 106Mhz keeps the roundtrip latency very low.

The latency will always achieve lower values than VDSL2 because of this fact for the operating frequency takes time at its lowest frequency; hence G.fast is very good for gaming.

G.fast roundtrip latency to BBC from Plymouth.

Latency for G.fast from Plymouth to London
Latency G.fast peaks are owing to SamKnows Whitebox performance testing

With G.fast I achieve 241Mbps/41Mbps synchronisation speed and 212Mbps/40Mbps throughput which was measured with a SamKnows Whitebox for ultrafast broadband on Zen internet. Clearly, I’m at ~180 metres from the PCP G.fast pod. DLM is still settling the line and has only been running for 18 hours. It will take action tonight for the line has been absolutely stable at 212Mbps throughput downstream and 40Mbps upstream on 30th January which is the first full day of operation.

Once 350 metres is exceeded by G.fast because of the higher frequencies its speed would be lower than VDSL2 that operates at lower frequencies and propagates further along the twisted pairs. 

For ensuring that G.fast does not interfere with the current frequency plan of VDSL2 the operating frequency is between 19Mhz to 106Mhz. Whereas VDSL2 uses frequencies from 0.025Mhz to 17.664Mhz to achieve its high speeds.

VDSL2 is pegged to a bandwidth of 80Mbps downstream and 20Mbps upstream at ranges to 400 metres from the cabinet, and at lower speeds at distances up to about 1.4km. There is no guarantee for speed on VDSL2 and this is especially true for upstream, but estimates are provided for downstream synchronisation speed for distance and line quality.

The symmetry for TDD in G.fast is mandatory 90/10 to 50/50 even though ISP’s only offer products for 300Mbps downstream and 50Mbps upstream at ranges to the cabinet of 150 metres, usually, and at lower speeds exceeding 150 metres to 350 metres maximum distance currently (practically <250m for an order); it has a guaranteed 100Mbps downstream threshold and 10Mbps upstream.

An estimate of between 141.7Mbps to 262.2Mbps, typical 231.8Mbps downstream and typical 17.8Mbps upstream were provided for my line on ordering G.fast for the top tier product for the ISP, Zen internet in this case. A router was provided by Zen internet the Fritz!Box 7530 but I’ve chosen to use my own an Airport Extreme 6th gen with PPPoE protocol to connect to Zen internet via the G.fast modem.

Observed speed on our G.fast connection
G.fast throughput speed observed