Re: New Gig1 Service Upstream Speeds are still poor
4 weeks ago
In the real world 940Mbps is fast enough for just about everyone not to be bothered if there is a reduction in speeds from an advertised 1104Mbps to 940Mbps. And anyone expecting a "gig" service of 1000Mbps should bear in mind that VM overprovisions Gig1 so you can get up to 1137Mbps at peak times, so there is some leeway if there is a slowdown. Also 940Mbps is quite adequate to handle 552Mbps (the minimum guaranteed download speed).
Re: New Gig1 Service Upstream Speeds are still poor
3 weeks ago
Until its mainstream on laptops and pc as standard no ISP is going to use it. If you require a specialist kits like £200 network cards or £500 routers. No customer would buy that or pay for the service if it had to include that. T
As a Very Insightful Person, I'm here to share my knowledge. I don't work for Virgin Media.
They aren't great - but then most offering the product are similar
3 weeks ago
I've a few minutes. I'm going to copy/paste some sections from posts on another forum along with paraphrasing earlier comments so that my own make more sense. The stuff in italics is comments others made, my own responses are underneath.
They've introduced 3.1 to download - why not upload? Even with no changes to the frequencies, moving from 256QAM to 1024QAM would give a 4x increase in upload bandwidth, turning that 50Mbps to 200Mbps. We’re not talking about maxing out the 3.1 standard here just enabling the standard in the first place.
And I do take your point that QAM1024 gets them pretty much to OR upload speeds. If it was that simple to implement QAM1024 then I suspect VM would have done it.
I suspect VM could implement QAM1024 in quite a few areas but there are other area where the physical plant just won’t cope.
Can anyone illuminate the upstream speed debate – my current understanding is that technically it is within the specifications to do so, but operationally the increased bandwidth will need the reverse transmission path equipment to be of sufficient capability since the downstream and upstream equipment (and I’m thinking amplifiers and repeaters) are different AFAIK
The upload currently runs at 64 QAM. That’s as high as 3.0 ATDMA goes.
1024 QAM doesn’t give quadruple the capacity 256 QAM does. It gives another 25%, at the expense of requiring 6 dB better MER.
Enabling 3.1 downstream doesn’t mean heavy duty upgrading of equipment – the line cards and CCAP already support it, it just needed configuration and licensing.
3.1 upstream can be enabled to run on the same frequencies as 3.0 by time sharing. A reasonable estimate is that it’ll allow for about 100 Mb upstream tiers.
This does require some care as it substantially increases load on return path lasers and risks clipping.
On the up side it doesn’t need replacement of amplifiers. Only perhaps some optical nodes that have Fabry-Perot optics with nodes with DFB diode lasers.
Gadget – it’s the same kit handling both forward and reverse paths. Diplex filters separate the upstream and downstream bands, they run through their respective amplification circuitry and are then transmitted in the appropriate direction.
A last point on upstream channels: the whole point of 3.1 is that it uses OFDMA and loads of small carriers. These will have various modulations depending on the MER of each carrier.
The SC-QAMs have 64 QAM as their target but can drop to 16 QAM or even QPSK if conditions require it. The 3.1 carriers can target 1024 QAM but most of them won’t get near that even with the much improved Low Density Parity Check error correction providing some coding gain over old school FEC.
One big bonus of the OFDMA return is that VM can go lower in the spectrum. At the moment they only go down as low as 22.6 MHz in most areas. An OFDMA block can go lower and tail off modulations as it reaches the noisier spectrum.
BTW the licencing is per MHz of channels for 3.1, which is worth considering. It’s per downstream or upstream channel on 3.0.
‘4.0 requires every node to be fed by fibre. ATM some of the nodes are daisy chained with heavy duty coax backhaul.’
All the nodes are fed by fibre. However they feed a series of coaxial amplifiers. The vast majority of cabinets in HFC areas house coaxial amplifiers. In the case of the about 500 premises passed node I’m typing this through there are, I think, nearly 20 coaxial cabinets.
There are 2 ways to do DoCSIS 4.0. One of them means no amplifiers while the other one does. One variety the upstream and downstream share the same frequencies and noise cancellation separates the respective signals, this is Full Duplex DoCSIS,the other they are frequency multiplexed so that can be run over amplifiers and doesn’t require fibre everywhere, that’s ESD – Extended Spectrum DoCSIS.
Feel free to share any of this on VM’s community forum. I haven’t used it in months and have no intention of doing so. <<< (Editor note: that aged well.)
Thanks Carl. What’s your interpretation on why 3.1 hasn’t been done upstream – from your comments it sounds more like penny pitching and lack of competition rather than any technical challenge?
Always comes down to money, Ferrocene.
They want to get Gig1 out to as many homes as quickly as possible. They’d have to upgrade the plant in some areas to allow for the increased return path laser load.
Depending on the kit they are using they might well have to upgrade return path line cards too.
They’ll go mixed mode eventually but for now the race is to get gigabit to homes before Openreach pass too many premises with the gigabit product they are releasing next February (I think?) or an altnet rock up.
Upstream behaviour on the Arris platform was…. interesting.
VM’s network architecture is not as open as OR’s or as consistent.
But all of this boils down to VM’s network can’t go a lot faster upstream without a decent amount of investment.
All I’m saying is that if you start spending that kind of money on upgrading the upstream pathway to QAM1024 to get a bit faster upgrading the whole thing to 4.0 might well be not that different in net cost and then you have a future upgrade pathway to boot.
This post is necessarily going to be technical. There is nothing in it that isn’t public knowledge. Feel free to share with credit anywhere you feel the need to 🙂
This first part is generic to cable companies as a whole and is just background on the technology.
On the matter of 1024 QAM: this comes with running OFDMA upstreams. The carriers are much smaller and some of them can and will run at much higher order modulations than the current SC (Single Channel) QAMs used in 3.0 and below.
Remember – the modulation order of a 6.4 MHz wide SC-QAM upstream has to be a lowest common denominator – a single spike of noise a few hundred kHz wide will prevent the channel running properly at a higher order modulation.
As I mentioned before a big part of the efficiency gain on OFDMA / DoCSIS 3.1 comes from being able to utilise more spectrum rather than the modulation order necessarily. Although it’s on Google for ease of access I will point out again that 1024 QAM is not quadruple the capacity of 256 QAM – the capacity is actually a function of the exponents of 2 – 4 QAM / QPSK = 2^2 = 2 bits per symbol / Hz. This carries on. 16 QAM = 2^4 = 4 bits per Hz, 64 QAM = 2^6 = 6 bits per Hz, 256 QAM 8 bits per Hz, 1024 QAM 10 bits per Hz.
With all these in mind jumping from the 64 QAM that VM currently use to 1024 QAM only increases the capacity by 2/3rds even if every single carrier runs at that 1024 QAM which isn’t going to happen due to impulse / narrow band noise.
Now, on the VM specific side.
For right now VM use up to 6 x 6.4 MHz wide, 64 QAM carriers upstream. They were previously bonding 4 of them per modem but have now started bonding all 6. There are reasons to bond 4 rather than 5 or 6 – there is a maximum transmit power range a modem can have been channels, can’t be too large between the lowest channel and the highest, and a maximum transmit power range in total depending on channel count being bonded as the transmitter in the modem can only transmit so many dBmV.
The ‘standard’ VM channel plan for DoCSIS 3.0 is this – I’m using the central frequency so the actual full channel extends 3.2 MHz in each direction up and down from here:
End to end that’s 22.6 – 63.5 MHz. Due to guard bands in between channels, can’t overlap them as you can OFDMA, roll-off, avoiding some impulse noise at certain levels there’s 40.9 MHz of bandwidth there of which 38.4 MHz is nominally being used by carriers and 30.72 million symbols of data are being carried a second.
OFDMA carriers can overlap ramp up and roll off, VM can go lower in the spectrum without any concerns over it getting noisier down there (power supply noise ripples up the spectrum, lower frequency = lower attenuation so ingress from outside is far more of an issue) and have no need to lose either the spectrum in between the SC-QAMs or the 1/6th of each SC-QAM that could be carrying data but is guard band / ramping.
No reason why they couldn’t stick a 48 MHz wide OFDMA band in between 63.5 MHz (I imagine they stop there to avoid amplifier roll-off in areas where the amplifier return paths stop at 65 MHz) and 15.5 MHz. With the higher efficiency, the coding gain from LDPC I mentioned earlier, higher average modulation order thanks to smaller carriers, having more RF bandwidth devoted to carrying actual useful symbols, etc, cable companies are reporting an 80%+ gain to capacity using the exact same spectrum as the SC-QAMs – VM can exceed this as, unlikely US cable companies with their profoundly limited spectrum, they haven’t had to go below 22 MHz – I’ve seen US cable companies going as low as about 14.
Which all thrown together makes a 10:1 ratio on the gigabit product realistic even in areas that are on the basic EuroDOCSIS 1.0 specification – 5 – 65 MHz upstream path.
Any networks that can’t meet that 5 – 65 MHz return path specification have/are being upgraded anyway due to both return path and forward path performance issues. They should be capable of 5 – 85 MHz return paths by the time that’s done, field upgradable to 5 – 204 MHz.
A last point on how open VM’s network is – while it’s not widely broadcast and the network capabilities vary they really vary in a similar manner to how Openreach’s copper network varies. Some bits of the Openreach network use different thicknesses of copper, some have SCPs connected to PCPs, etc.
The VM networks are hybrid/MDU HFC cascades. An optical node has a certain number of coaxial branches it powers. These branches split into smaller branches which happens in cabinets. In most of these cabinets there are coaxial amplifiers and tap banks. The signal gets amplified and split to be fed into a tap bank on one split where it goes into properties and to go to the next amplifier in the cascade.
There might be exceptions but this covers the overwhelming majority of architectures. The capabilities and performance of the individual optical nodes and amplifiers may vary but it’s that basic infrastructure so it’s fairly well understood.
The other varieties are trunk and branch style, these tend to be overground on poles where amplifiers are placed periodically on trunk lines and are tapped wherever they need to be to feed homes, and star networks where an optical node has a series of coaxial lines coming out of it each feeding an amplifier which then feeds homes.
It’s really tricky to explain it words, sorry if it doesn’t make sense.
Commercials: Basically once BT / A N Other with big network reach make a big deal out of upstream speeds. At the moment and since time immemorial they don’t. Once that killer app for upstream speeds appears, is adopted, and makes a catchy selling point then VM start turning mixed-mode on.
I would imagine they’ll also be handing Hub 4s to people not on Gig1, too, to take load off the old SC-QAMs and 3.0 downstreams. While they are quite a bit more expensive than the Hub 3 it’s also very expensive and commercially silly to be paying for the licences and infrastructure to run DoCSIS 3.1 and having a small fraction of the customer base using it, while the 3.0 network’s utilisation increases constantly.
Interesting if the Hub 4 is a lot more expensive why it hasn’t been future proofed. No 802.11ax and sticking to 1GbE ports mean that it’s useless when the next generation comes along… I guess they’re confident that there won’t be a need for a package above 1Gbps for a long time.
Not sure how much it would have cost to put a 2.5/5/10G NIC in. Existing backplane capacity should be fine (at least with a 2.5/5 port).
Based on what you say they can migrate lower package users to it but VM will be reluctant to do so. Looks almost like the CPE was rushed to the market.
That CPE is the same one 3.1 was released on last year in other Liberty Global territories. The extra cost is from the 3.1 modem. 3.1 silicon is still not cheap though the cost is slowly coming down.
No real business case for multi-gig ports and definitely none for 802.11ax – how many people have multi-gig NICs or 802.11ax cards? Remember 802.11ax hasn’t been finalised yet, it’s still in draft phase.
Later on they can release CPE with multi-gig ports and 802.11ax if they need to but for now the uptake of even the Gig1 service is going to be so low for a while that there’s no real point in incorporating them into this first incarnation.
VM have 2 varieties of Superhub 2 – the standard and one with 802.11ac WiFi. The group may elect to release an updated version if/when products of >1 Gb become required.
Certainly wasn’t rushed to market – it’s the standard 3.1 CPE across Europe Liberty Global are using and have been for a year.