Kitz Forum

Broadband Related => Broadband Technology => Topic started by: Weaver on June 01, 2018, 03:08:42 AM

Title: Better DSL modems
Post by: Weaver on June 01, 2018, 03:08:42 AM
From what I have just been reading (https://www.osapublishing.org/viewmedia.cfm?uri=oe-22-9-10844&seq=0) about non-rectangular QAM constellations (https://en.wikipedia.org/wiki/Quadrature_amplitude_modulation#Non-rectangular_QAM), it does seem that modem design can be pushed quite a bit harder still in order to get more speed over any line, fast links or even over very long slow lines. More complex modems can be built as transistor counts keep going up, and non-rectangular optimal QAM devices need very high speed complex circuitry because they have a harder time decoding the symbols at the lowest physical level. The paper I read discussed a number of options but one was base 31 symbol system, not binary, each symbol had one of 31 possible values, so you have to convert binary to base 31. Thatís easy enough if you just accumulate 31 * 32 bits in a buffer, so 31 * 4 = 124 bytes come in and you transmit 32 symbols. The constellation was based on hexagonal packing, the best packing there is, so that is the reason for the funny number. The system to decode or generate those coordinates on that constellation is not simple like normal rectangular QAM. It was for a 100Gb optical laser system and it suffers from bad non-linearity problems apparently. Predicted to use around 1 W.

Their choice of what is Ďoptimalí was to even out the probability of error across the constellation so that the worst-case minimum euclidean distance was a minimum, and so no one case was unduly vulnerable to error. This is the right thing. Other articles have discussed minimising the maximum transmit power only, but that is not enough. Sorting the error rate out means you can push the whole thing harder and get more speed. The circular or nearly so hexagonal constellations minimise the worst-case / maximum enclosing radius anyway.

The thing had a much better bit error rate in a simulator than normal DMT that we are used to. I would need to see the numbers for much much larger constellation sizes though to make a better comparison with DSL as we are used to it.

Unfortunately, I think it is perhaps too late for the likes of me as upgrading the PHY of modems at both ATU-C and ATU-R to bring in a new standard would be expensive and service providers are not keen on spending money.

[Moderator edited to fix a broken link.]
Title: Re: Better DSL modems
Post by: burakkucat on June 01, 2018, 05:21:28 PM
I have, only just recently, managed to understand a little about rectangular QAM and the efficiency gains thereof. Now you mention research into (approaching) circular constellations . . .  :swoon:

For your amusement, I will point you towards a five page thread (https://forum.kitz.co.uk/index.php/topic,11293.msg) with the title "Broadcom DSL Diagnostics App", that was started six years ago.
Title: Re: Better DSL modems
Post by: spring on June 02, 2018, 10:49:26 PM
On my VMG1312-B10A in modem-router mode with v14 firmware, it made it reboot after what seems to have been a crash of its software, after Broadcom DSL Diagnostics tool couldn't connect due to a socket error.
Edit: Tried entering "dsldiagd" in BusyBox and confirmed it worked with "netstat -lu" listening on UDP 5100 but still got a socket error & modem reboot. I don't have SSH/SNMP on the device, just telnet, maybe that's it. Even though I won't try another time I hope this post is informant.