Recently, I have been bedeviled by unstable Internet service from COMCAST (Xfinity) in the Boston area. The following story is not limited to COMCAST, and it could have been any Cable-TV Internet Service Provider.
Starting in late July 2021 and ending in early August 2021, for about one month total, my Internet service became very unstable. (Cable TV service was unaffected, so it was not that the physical cable path was interrupted.) Internet service had been rock solid since about 2016, when I purchased and installed the current cable modem, an ARRIS SB6183. My service tier is 200 Mbit/sec download, which is well within the capabilities of the SB6183, 686 Mbits/sec.
The main external symptoms were that browser access would randomly slow and hang, sometimes recovering, something not. Likewise email (POP account). Rebooting the modem (by cycling power) often but not always restored Internet service. As the month progressed, things got progressively worse, although it was always relapsing-remitting - it should spend all night trying to connect, failing, trying again, and so on.
Called COMCAST tech support, which forced me to deal with a robot lady that could only follow a script, the main point of which was to get me to power-cycle the modem. This power cycle often did work, probably because she was probably also sending a modem reset signal at the same time, but she never made that clear. (The robot lady here is in fact a machine, not just an unimaginative human, a droid.)
Got a human later, after calling back and answering that the problem was not fixed. The human came to the conclusion that my cable modem was too old and likely broken, and insisted that I call ARRIS and have them diagnose the modem, and only if ARRIS declared that the modem was not broken would COMCAST do anything like send a truck without charging an arm and a leg.
Called ARRIS. Cost US $50 (out of warrantee). Got a real human, in Chennai, India, who did seem expert in the modem internals, with perfect English and a very good telephone line, and was very patient.
COMCAST had claimed that ARRIS could do technical (ie, electrical) tests remotely, but this turned out not to be true. Anyway, I had collected lots of status data from the modem, which we discussed, and he concluded that the modem was not broken and did not need to be replaced
Now one mystery had always nagged me: While the Internet was up, the modem would report received signal powers (over 16 parallel RX channels) of +5 dBmV (decibels over one millivolt in a 75-ohm system), and a SNR of about 40 dB, and yet errors (both corrected and uncorrected) kept accumulating.
When the link was down, the received signal level would drop by 20 dB to -15 dBmV (which is in the DOCSIS 3.0 Spec Range) , and the SNR would still be 35 dB (also in spec range).
How does that work? With 35 to 40 dB SNR, there should be no errors. Neither COMCAST nor ARRIS were able to interpret that oddity.
Called COMCAST back. Went through the robot lady yet again, but on the third call, got another real human.
He really had no idea what went on inside modems et al, and insisted that if the associated firewall/router was not also rebooted when the modem was reset or rebooted, the ensuing chaos would cause the modem to be unable to measure incoming RF power (at around 500 MHz) correctly on a coax, the claim being that both Ethernet and Cable TV were both "electrical", so they could heavily affect one another Hmm. I'll have to think about that. For a very long time.
He was also of the opinion that Ethernet wires wear out and need periodic replacement. I declined to disassemble the house to get at the CAT5e cable that runs from basement (at the modem) to the 2nd floor (where the WiFi base station lives).
He worried greatly about cables and connectors in general, saying that they were very often the cause of such problems as I was seeing. After one hour discussing the issue, while his theories of causation were nonsense, this was his direct experience, and thus was my main takeaway.
Usually, connectors are the main cause of problems with "cables", and it had to be a coax cable, so I found an 11mm (7/16") open-end ignition wrench, and went around loosening and re-tightening all CATV RF connectors (Type F, for the record). The loosen-then-tighten drill is to physically disrupt any corrosion at contact points.
All connections were tight save one, the one where the coax from the modem connects to the RF feed coming from the nearby boiler room. The loose connector is in the wall plate under the table upon which the modem sits, along with my desktop computer, and that connector had managed over the years since 2016 to unscrew itself almost to the point of falling apart. That cable does move a bit when things are added or removed, and so on. Tightened the connector. Bingo!
And now the cause of the oddity became clear. The SNR is measured over time, to yield a stable value. But the connection was chattering between normal power (+5 dBmV) and open (-15 dBmV) fast enough that the SNR calculation didn't notice. This was also too fast for AGC et al to react, so symbols sent during the open period were lost. Some could be recreated by forward error correction, and some could not, and these errors accumulated steadily over time despite the stellar SNR.
The modulation type is QAM256, and the symbol rate on each QAM256 channel is 5120 Ksymbols per second. There are 16 downlink and 4 uplink channels. The AM component of the QAM symbols will have a fundamental at 5 MHz or so, far above the spectrum due to a rattling connector.
.
A spectrum analyzer would not have found this problem either. I suppose if one envelope demodulated the received RF, the chattering would have been quite obvious - big AM noise peak in the 10 to 1000 Hz range, or the like. I would think that it would be useful for the modem to do this - loose connectors are pretty common I'd hazard.
Anyway, it all works now, the key symptom being that there are no new errors accumulating now.
And the takeaway is that if SNR exceeds 30 dB and yet errors accumulate, start re-tightening RF connectors. If that doesn't work, look for a cracked wire or connector, or a loose shield.
Joe