I decided on the DPC3010 because Cox has more frequent firmware updates and better hardware support for it. The CMTS is after all a Cisco.
Im a EE and I constantly improve any device I use. Right off a large number of things in the design stood out as needing improvement. In fact there were a number of spots on the PCB in the power supply that were not populated with parts. This was a obvious cost saving step and they did just enough in parts to squeak by Cable Labs certification. It was clear a lot could be done to improve the hardware.
A Cable Modem is a very analog device. QAM is a highly complex and fairly finicky system. The cable modem transmits a complex signal that must be incredibly timed and very precise. While it has systems for error correction these error corrections cause some cpu time and produce jitter. Uncorrectable errors produce a retransmission and a larger jitter component. How good the sync is with the CMTS also effects jitter.
Jitter is a issue that effects some key things. Like gaming or to some degree VoiP. Obviously less correctable errors the better. There are other measures of a connection I wont bore you with here but all are effected by the analog performance of the cable modem.
Right away the power rails of the CM were obviously hurt by cost reductions in manufacturing. I stuffed a huge number of huge solid tantalum caps across these rails, in each switching power supply for each of the 3 power rails, put them at the points on the BCM3380 where power enters the chip and also addressed the special power rail that supplies the LNA chip in the RF section. Overall it was a hugh amount of tantalum. I think I dropped in twice the cost of the whole modem just in tantalum caps. I also used some big electrolytics to provide the big guns to quiten the supplies all the way down. I used tantalum for a number of reasons but low ESR at high frequencies is the most important here.
The overall effect on the power rails was dramatic. Measured at the chip power entry pins a 1,000:1 reduction of noise occurred. Also the chip now had very stable power rails to create the upstream signal and very very quite rails to listen to the downstream QAM.
The performance effect was interesting. A 1db increase in downstream levels occurred along with a full 1db increase in S/N. The upstream power decreased by 1.5db indicating that the transmitted signal was being seen more clearly now. My correctable errors dropped by half. My buddy at Cox told me all my performance numbers had a dramatic increase in performance.
My levels ended up at +6-7 downstream and 38.5 up with a 41.1 db S/N in a 8x4 config.
I purchased 2 DPC3010's. I kept one unmodded to compare.
I used Multiping set to do 10 pings a second and chart it. My target was the CMTS IP. I did weeks of charting and then did mods and did more charting. The differences were quite interesting.
The large jitter was reduced very obviously. A much smoother and consistent connection resulted.
I have the Ultimate Tier of service from Cox and in my area its capped currently at D 65 U 5.5. The mods did not change this but testing did become VERY consistent.
Of course other things improved. The stability of the modem was seriously improved as the supplies were now noise free and the regulation was very consistent. Impulse current draw by the CPU is easily fed from the caps. I expect one of the big performance improvements is long term reliability.
Speaking of reliability...
The thermal management of both the SB6180 and DPC3010 is PISS POOR. These things are way overheated. While it might not be a problem initially over time the high heat will degrade performance and eventually kill the modem. The heat sink was crap. I replaced it with a much better heat sink, using Artic Silver heat sink compound and added a Silenx fan. I also modded the case to allow far more air thru the case. I power the fan off the 12V power coming into the unit.
Speaking of the power supply. Because of the higher inrush current caused by the startup of all the new caps a larger power supply was needed. I settled on a 3A 12 supply. This is overkill but the .4 Adc nominal that the modem draws wont even heat up the 3.5 amp supply again providing long term reliability improvements as the power supply is not gonna burn out any time soon.
A also added a TII gas based surge suppressor in the coax right before the modem to protect it from the outside world. http://tiinetworktechnologies.com/repos ... 6-0710.pdf
The caps used were:
The yellow ones are 100uF 16V kemet Tantalum
The black caps to the left of the heat sink are 1000uF 16 v and 2 are stacked per rail.
I used a 560uF wet tantalum - its the silver one. Its highly expensive and I used it as I had it lying around. You dont really need it.
There is a big 3300uF 16V electrolytic on the right of the heatsink.
Some pics...
A bit fugly, but functional.
It was a fun project and I think I have the best DOCSIS 3 modem on earth ! hahaha...
The improvements are noticable using the web. I think gaming would be better as there is less jitter. It would be more responsive. For normal web browsing it might seem a bit more response and quicker.
I have a fairly fast home networking setup. Im using a older gaming machine running Vyatta which turns it into a Linux based high power router. I am also running a high performance rDNS caching server on it with settings for performance and carefully chosen DNS peers based on performance. All my wiring is cat 7a shielded.
My electronic test equipment list is extensive and I did lots of measurements as I went in the mods. I have everything from a 1Ghz spectrum analyzer to great scopes, rubidium based freq counter and great DMMs.
I figured you guys are already doing tweaks to cable modems so this might be a fun project. I posted here as you guys understand that the problem is, just like other endevors on this site, modifying a cable modem is a no no...
Have fun !
