2
\$\begingroup\$

The problem:

The ADSL2+ service at home is not performing up to its potential. The modem is supposed to sync at 16 Mbps (downstream) but the it does only up to 11.5 Mbps. Using the diagnostic tools provided by the modem, I figured out there's noise in a particular frequency band that renders that band unusable by the modem. I need help in identifying the source of this noise.

Background info:

ADSL2+ service uses the frequency band from 140 Khz up to 2.2 Mhz for downstream. It uses carriers ("bins") that are 4.3125 KHz apart. Each bin can carry up to 15 bits of data. How the total capacity is allocated to bits in each carrier is decided by the modem. The better the SNR for a certain carrier, more bits are allocated.

The issue:

The modem that I have (Vigor 2820) has a diagnostic utility through its telnet interface that tells how many bits are allocated to each bin. In my case, it is showing that frequencies from 1.737 Mhz to 2.087 Mhz are unusable. Interestingly, frequencies from 2.087 Mhz up to the maximum 2.2 Mhz are used by the modem.

Normally, bit distribution is dynamic based on line noise etc, but in my case those frequencies are never used and the modem cannot sync at the 16 Mbps promised by the ISP. I understand that bad cabling and similar problems may exist but since only a particular frequency band is effected, I am not sure bad cabling is the problem here.

The crow's fly distance to the CO (central office) is 1.58 km, which according to https://www.increasebroadbandspeed.co.uk/2012/graph-ADSL-speed-versus-distance means I should be able to get about 19.7 Mbps with 22 dB line loss. In my case, I get 11.5 Mbps with 34 dB line loss. I am aware that the actual wire run may be more than the crow's fly distance.

Please note that this is an EE question and not a "how to fix my ADSL connection" question. I am really after finding out what kind of a noise source renders this particular band of frequencies unusable.

What I did so far:

First, here's how the situation looks if I force the modem to sync at ADSL2 speeds only (i.e. use only up to 1.1 Mhz). I do not use this protocol normally, but I included it for completeness.

With ADSL2 only

In this picture, x axis is the frequency and y axis is the number of bits allocated at that carrier frequency. For those with ADSL service and Vigor-brand modems, you can try the same with your service. Connect to your modem's telnet interface and copy/paste the result of "adsl showbins" and "adsl showbins up" commands to https://sdfjkl.org/hack/dmt-visualizer/dmt.html

Second picture shows how it is during normal usage. The unusable frequency band is obvious.

With ADSL2+

Thinking that it may be noise from the switch-mode power supply, I fed the modem 12V from a lead-acid battery. Third picture shows this case. There's a bit of a general improvement in bit distribution, but those frequencies are still not used.

Running on batteries

Remembering that I have ethernet-over-power adapters at home, I checked the frequency band those adapters are using. It turns out HomePlug AV standard uses frequencies from 1.8 to 30 Mhz. Thinking that this somehow may effect the modem's power supply, I took them off and rechecked the bit distribution. Unfortunately, not much changed:

No Powerline devices

I also thought "the modem is a bit old so maybe it is the modem itself", so I switched to another modem (Vigor 2860) with its own power supply. Results were still similar.

Next step was using an oscilloscope (Rigol DS1054Z) and its FFT functionality to see if there was any obvious noise in that band (I was not after ADSL carriers or anything like that). I did not see anything obvious in the FFT. Either there was really nothing obvious in that range or I could not use the scope properly (I am a newbie when it comes to digital oscilloscopes and this brand in particular).

The questions:

1- Does this frequency ring a bell to anyone? It is outside of AM transmission so that can't be it (not to mention all wires are underground until the building entrance). I found a frequency allocation table at https://www.ecodocdb.dk/download/2ca5fcbd-4090/ERCREP025.pdf but it wasn't very helpful in identifying what might be in that range.

2- Would bad connections or corroded wires cause such frequency-specific interference?

3- Within the building and the unit, the phone cable runs next to the cable-TV cable (for about 40 meters or so). Would this somehow effect the ADSL signal?

UPDATE:

Based on the answers and questions, here are more details:

The cuts in the cable run from CO to our unit is like the following:

CO--->Curbside cabinet--->Building Demarcation box--->Our unit entrance--->Room with the modem, computers etc

CO--->Curbside cabinet is about 1.58km (crow's fly), Curbside cabinet--->Building Demarcation box is about 50 meters, Building Demarcation box--->Our unit entrance is about 20 meters and Our unit entrance--->Room is about 20 meters. The curbside cabinet is passive, about 35 years old. The same goes for the demarcation box. It's supposed to be locked but in our case it's not and according to the ISP, we have the right to do maintenance work in it after their approval (I assume there's a "client" side to that box that we are responsible for).

BTW, this is a high-rise apartment building with about 20 units. We live somewhere in the middle.

Since we are talking about impedance mismatch, I checked up on what sort of wires are used in the run. Apparently wire gauges are not necessarily uniform from CO to the customer premises according to https://www.speedguide.net/faq/how-does-signal-attenuation-relate-to-wire-gauge-373 Since I do not have access to the curbside cabinet, I do not know the actual gauge of the wire until that point. But I measured the wires that come in our unit (pair of very thin bare wires) with a Mitutoyo digital caliper and the diameter turned out to be 0.32 mm, which I think corresponds to 28 AWG. The in-unit run was replaced a while ago with CAT-5. I measured that CAT-5 cable to be 0.34 mm, which I assume is out of spec (cheap knock-off cable I assume --- I do not even know if it is copper or copper-clad aluminum). Maybe I should replace that so-called "CAT-5" cable with something better. I have some CAT-3 cable that I measured to be 0.45mm. I do not know which one is better in terms of impedance match.

As for the location of DSL filters and branches in the run: There aren't any. The cable goes straight into the modem. I have a separate line for voice.

About CB and amateur radio: I looked into that and CB seems to run at about 27 Mhz which doesn't seem to be a concern. Also, the only frequency band licensed locally for amateur radio is from 1.81 to 1.85 Mhz. I am not sure if there are any towers nearby. Would this effect the whole range I am having problem?

I disconnected the cable-TV cable but it did not make a difference.

I will move the modem close to the unit entrance and do more experiments but that will require some undisturbed free time on my side and a quiet time in the family (i.e. nobody using the internet). I will update once I am done.

What I did, however, is to remove the small box at the unit entrance and just twist the wires together. Now the modem syncs at ~11.7Mbps, the line loss is reported as 33 dB instead of 34 dB. That box looked like the following:

Box at unit entrance

I can't attach a picture of the bin distribution using the link I used before because now the modem reports bogus numbers for bit distribution and that software just doesn't run.

Here's an example. This one is from when the numbers are reported correctly:

-----------------------------------------------------------------------------
Bin  SNR  Gain Bi - Bin  SNR  Gain Bi - Bin  SNR  Gain Bi - Bin  SNR  Gain Bi
      dB   dB  ts         dB   dB  ts         dB   dB  ts         dB   dB  ts
--- ----- ---- -- - --- ----- ---- -- - --- ----- ---- -- - --- ----- ---- --
  0     0    0  0 *   1     0    0  0 *   2     0    0  0 *   3     0    0  0
  4     0    0  0 *   5     0    0  0 *   6     0    0  4 *   7     0    0  0
  8     0    0  0 *   9     0    0  0 *  10     0    0  0 *  11     0    0  0
 12     0    0  0 *  13     0    0  0 *  14     0    0  0 *  15     0    0  0
 16     0    0  0 *  17     0    0  0 *  18     0    0  0 *  19     0    0  0
 20     0    0  0 *  21     0    0  0 *  22     0    0  0 *  23     0    0  0
 24     0    0  0 *  25     0    0  0 *  26     0    0  0 *  27     0    0  0
 28     0    0  0 *  29     0    0  0 *  30     0    0  0 *  31     0    0  0
 32     0    0  0 *  33     0    0 12 *  34     0    0 12 *  35     0    0 12
 36     0    0 13 *  37     0    0 13 *  38     0    0 13 *  39     0    0 13
 40     0    0 13 *  41     0    0 13 *  42     0    0 13 *  43     0    0 13
 44     0    0 13 *  45     0    0 13 *  46     0    0 13 *  47     0    0 13
 48     0    0 13 *  49     0    0 13 *  50     0    0 13 *  51     0    0 13
 52     0    0 13 *  53     0    0 13 *  54     0    0 13 *  55     0    0 12
 56     0    0 12 *  57     0    0 12 *  58     0    0 12 *  59     0    0 12
 60     0    0 12 *  61     0    0 12 *  62     0    0 12 *  63     0    0 12
 64     0    0 12 *  65     0    0 12 *  66     0    0 12 *  67     0    0 12
 68     0    0 12 *  69     0    0 12 *  70     0    0 12 *  71     0    0 12
 72     0    0 12 *  73     0    0 12 *  74     0    0 12 *  75     0    0 12
 76     0    0 12 *  77     0    0 12 *  78     0    0 12 *  79     0    0 12
 80     0    0 11 *  81     0    0 11 *  82     0    0 12 *  83     0    0 11

Notice that bits per bin is no more than 15.

This one is from when the numbers are whacked:

-----------------------------------------------------------------------------
Bin  SNR  Gain Bi - Bin  SNR  Gain Bi - Bin  SNR  Gain Bi - Bin  SNR  Gain Bi
      dB   dB  ts         dB   dB  ts         dB   dB  ts         dB   dB  ts
--- ----- ---- -- - --- ----- ---- -- - --- ----- ---- -- - --- ----- ---- --
  0     0    0  0 *   1     0    0  0 *   2     0    0  0 *   3     0    0  0
  4     0    0  0 *   5     0    0  0 *   6     0    0  0 *   7     0    0  0
  8     0    0  0 *   9     0    0  0 *  10     0    0  0 *  11     0    0  0
 12     0    0  0 *  13     0    0  0 *  14     0    0  0 *  15     0    0  0
 16     0    0  0 *  17     0    0  0 *  18     0    0  0 *  19     0    0  0
 20     0    0  0 *  21     0    0  0 *  22     0    0  0 *  23     0    0  0
 24     0    0  0 *  25     0    0  0 *  26     0    0  0 *  27     0    0  0
 28     0    0  0 *  29     0    0  0 *  30     0    0  0 *  31     0    0  0
 32     0    0  0 *  33     0    0 11 *  34     0    0 12 *  35     0    0 12
 36     0    0 12 *  37     0    0 12 *  38     0    0 12 *  39     0    0 12
 40     0    0 13 *  41     0    0 13 *  42     0    0 13 *  43     0    0 13
 44     0    0 13 *  45     0    0 13 *  46     0    0 13 *  47     0    0 13
 48     0    0 13 *  49     0    0 13 *  50     0    0 13 *  51     0    0 13
 52     0    0 12 *  53     0    0 12 *  54     0    0 12 *  55     0    0 12
 56     0    0 12 *  57     0    0  0 *  58     0    0  4 *  59     0    0  0
 60     0    0 115 *  61     0    0 129 *  62     0    0 132 *  63     0    0 110
 64     0    0 12 *  65     0    0 12 *  66     0    0  4 *  67     0    0  0
 68     0    0 115 *  69     0    0 129 *  70     0    0 196 *  71     0    0 110
 72     0    0 12 *  73     0    0 12 *  74     0    0 12 *  75     0    0 12
 76     0    0 12 *  77     0    0 12 *  78     0    0 12 *  79     0    0 12
 80     0    0 11 *  81     0    0 12 *  82     0    0 11 *  83     0    0 12

In this case, bin 60 has 115 bits, bin 61 has 129 bits, bin 62 has 132 bits, bin 63 has 110 bits etc. Either this is a firmware bug or something is happening in bins 60-71 (258.75-306.1875 Khz).

Since the initial software is not running in presence of bogus data, I switched to one called OrbMT. It can deal with out-of-range numbers and reports a bit more data. I attached a screenshot below.

OrbMT screenshot

My "twist the wires" action seem to have one positive effect: It brought back bin 470 (2026.875 KHz). Yay!!!

UPDATE 2:

Here's what I did based on recent feedback:

I tried the 100 ohm-1000pF termination suggestion from @analogsystemsrf but that did not help with the gap. I attached a screenshot showing the result.

With termination

Then I removed the termination and stripped the wires a bit more and tying them together with more twists. The result is below.

More twists

Apart from the speed increase, one noticeable change is that the CO started broadcasting with more power (18.2 dBm instead of 17.9 dBm).

Next experiment involved using a POTS cable of similar length instead of the CAT-5 cable in the last run. Compared to the CAT-5, the connection rate got lower, transmit power got lower, and less bits are used in the higher frequencies after the gap.

With phone cable

Before I put everything back to their original state, I stripped the wires even more and made more twists. Here's the result:

More twists in the wire

This is the best speed I could get recently, with the CO broadcasting at 18.4 dBm.

@Edgar Brown suggested the proper diagnosis could be made by using time-domain reflectometry and included a link that shows how to make one with an oscilloscope and signal generator. That link is very useful as a starter but it does not discuss the proper modifications for live circuits. The signal generator I have is a cheap one and I do not know if it would withstand the voltages on the phone line. In any case, I will study it using the cables that I have and learn about the subject. I will also see if using any of the cables that I have will fare better for the last cut. I will update once I find more.

\$\endgroup\$
  • \$\begingroup\$ Go outside, and disconnect the cable-TV. How long has this gap existed? \$\endgroup\$ – analogsystemsrf Feb 7 at 20:06
  • \$\begingroup\$ Does your modem give you gain or attenuation by frequency? That a better thing to plot, as it'll indicate if its the line that's bad, or if your provider has just decided they'd rather not use that frequency. If you think it's interference see if it's time dependant. Also you can use a short wave radio to tuned to that band. The usual suspects are cheap halogen light "transformers" bad wall warts. \$\endgroup\$ – james Feb 8 at 0:15
  • \$\begingroup\$ @analogsystemsrf: I disconnected the cable-TV but it did not change anything. I am not sure about the gap's existence. I seem to remember this line synced at above 12 Mbps at one point but I am not sure. What I am sure, however, is that the upload used to sync at exactly 800 Kbps (ISP cap) but now it syncs at 796 Kbps. So something has happened. \$\endgroup\$ – alokoko Feb 8 at 19:11
  • \$\begingroup\$ @james: Unfortunately this particular model doesn't give that information. That number is always reported as 0. I am keeping an eye on that gap now using the OrbMT software. I do not have any halogen lights. Should I worry about all the wall warts in the unit? \$\endgroup\$ – alokoko Feb 8 at 19:13
2
\$\begingroup\$

One possibility that you don't seem to have considered is a resonant segment of wiring that is acting as a filter.

A ~100m segment with enough of an impedance discontinuity could create reflections that reinforce each other while a similar ~50m segment could create reflections that cancel each other at those frequencies. If such reflections are large enough (too large of an impedance discontinuity) the ADSL equalization algorithms might not be able to compensate for them.

Given that this is the approximate run length within your house I would suspect a large impedance mismatch where the phone line enters the house (perhaps it branches out into multiple rooms?).

I would repeat the tests but tying the cable modem directly at the house entrance point and disconnecting the rest of the house.

If that fixes the problem, consider installing an ADSL line filter at the entrance point and route a dedicated tap to be used just for the modem.


POTS cable and CAT5e cable are likely to have different impedances in the range of interest. Perhaps not different enough to matter, but it is a possible issue. This question has some relevant information and suggests that, at least in some circles (perhaps without justification), CAT5 cable is not seen as a good alternative for ADSL.

If the impedance mismatch is an issue, and there is enough additional SNR in the link, adding a small amount of series resistance or a parallel AC-coupled resistance at the connection point (as @analogsystemsrf suggested) could reduce the mismatch and introduce enough attenuation to any undesired resonance.

The proper way to know for sure is to use time-domain reflectometry in the link. At these frequencies and distances you can put one together with an oscilloscope and a signal generator (or even a micro controller with either a fast ADC or a couple of comparators).

\$\endgroup\$
  • \$\begingroup\$ Hang 100 ohm in series with 1,000pF as a non-DC_termination across the ADSL line. \$\endgroup\$ – analogsystemsrf Feb 7 at 20:30
  • \$\begingroup\$ No branches and no phones. Impedance discontinuity is looking like a real possibility. I made some updates to my post and will continue as I find more. \$\endgroup\$ – alokoko Feb 8 at 19:15
  • \$\begingroup\$ @alokoko added some relevant information to the answer. \$\endgroup\$ – Edgar Brown Feb 8 at 20:07
  • \$\begingroup\$ Thank you for the link. I will study that and experiment with various cables that I have. I am not sure how to modify the setup so that it is safe to use in a live ADSL circuit. In particular, I am not sure whether my cheap signal generator will take the voltages on the phone line without any issues. \$\endgroup\$ – alokoko Feb 10 at 11:48
  • \$\begingroup\$ @analogsystemsrf: I did that but very little changed. In particular, the gap is still there. I made some edits to my post (update 2) with my findings. \$\endgroup\$ – alokoko Feb 10 at 11:49
2
\$\begingroup\$

Update 3 : What’s all this stuff about RL, N bits/baud vs SNR and CMR

All (old baseband and modern ADSL, cable ) modems with N bits/Baud can handle impedance discontinuities. Some more than others. They also use superior balanced impedance (baluns) in a 2-way “hybrid” transformer.

Even old 56kbps modems which used 4kHz audio in full duplex (FDX) with N=8 which could support 64kbps but only supply 56kBd due to aggregate of fringe EMI egress rejected by the FCC in the late 80’s~90’s.

Software Modems were cheap and handle only 1 impedance discontinuity or echo , would throttle down while better DSP types could achieve full bit rate. These were the best choices. All audio SLICs at the CO active pre-emphasis to equalize the gain for each Subscriber. Longest lines @1 mile= 1.6km used hybrid 2way negative impedance line amps that had to be carefully tweaked with extra efforts to raise SNR in the 50 dB range. The loop gain had to be adjust just <1 to avoid oscillations which sounded like spurious white noise to clicks.

If the Common mode noise or crosstalk or return loss or quiet termination (secret phone #) did not deliver adequate performance, the max N bits per channel could not be used.

Now, the better home DSP modems could handle 2 or 3 echoes over long-haul Return Loss link errors, one could get one echo delay from each badly tuned CO to CO link which used DS0 64kb/s mu or A-Law digital channels in between as part of a DS1 1.544 Mbps or in EU 2.048 Mbps (because they started later with current technology developed in USA and Canada).

The attenuated amplitude echo is called Return Loss. (RL or s11,s22) There were many audio test patterns used to phase lock frequency, phase s-parameters in full Duplex while converting async 10bit UARTs with start/stop to 8 bit sync , stripping off the start-stop bits. Fax modems also used is technology but users often paid extra for a special line and SLIC tuning while some other could get it free like now. (Even though Fax is almost obsolete except for really old-school lawyers and insurance companies) Some DOS S/W even mis-reported 125% performance just because of that conversion of 10 to 8 bits)

The old audio 56k modems performed an Inverse Fourier Transform after an initial linear and fixed (1st and 0 order correction) sync at 1200/2400 Hz for Group Delay , phase error null, then a PRSG white noise was used lastly with muted audio to fine tune the spectral response to correct or “flatten” the cable response to try to achieve max 8bit/baud then throttle down if not.

Fast forward 30yrs and ADSL2+ can now get up to 15 bits per baud per roughly 4kHz Band just they need 20log N more SNR to achieve this while rejecting CM conducted noise, CM and DM radiated crosstalk, DM noise, then phase and group delay errors from changes in DM and CM impedance, ...

... then we have Ham guys who interfere or megawatt urban train traction motors and broadcast radio stations that must be attenuated by almost perfectly balanced impedance. Also since a 1st order filter starts phase shift more than 2 freq. decades away, there is still a couple degrees of phase shift or <1% in a LPF like line length with a 1/4 stub at 5 or 10MHz having an influence at 2MHz or down to 100 kHz.

So with the number of multi pole-zero corrections even on a old DSP modern means your not going to get 15 bits/baud with a couple impedance discontinuities. It needs much tighter flatness and much higher SNR to achieve full data rate, which is possible << 1 mile but any of the above kinks or ingress etc that are the “hair on the camel’s back” may be too much for 1+ mile.

I’m have not seen ADSL2+ modem designs but I suspect they use best FIR filters with some of the power of a Network Analyzer to calibrated all 8 s parameters for each band, so it takes a minute or so. But the same fundamental issues exist from RF ingress like traction motor trains and Ham operators then CMR of crosstalk from other Subscribers and Hybrid transformer balance of 50 dB with wide bandwidth is really hard to achieve in magnetics if not impossible so state-of-the-art digital compensation is used.

So what can be improved?

Cable Impedance is not just wire diameter but the ratio of wire diameter to dielectric thickness or \$\sqrt{L/C}=Zo\$ that controls impedance. This also is affected by number of twists per “ or Cm.

  • Try a large low permeability ferrite magnetic Baluns for entire RF BW to improve CMR (common mode rejection)

  • AC Pi filters? to shunt some CM noise to earth gnd while raising DM impedance and minimizing affects on much lower DM impedance.

  • Better DC ext power “wallwarts” with lower ripple 12V power using lower ESR caps to replace aged ones??

  • Use a Network Analyzer or TDR to test the link?

  • Use the modems internal diagnostics to report link SNR and noise type.?

    ————————————————$—————————-#$&@*%—————————-

I would imagine a network of unterminated jacks in the home would cause spectral notches at 1/4 wavelengths around the home unless there is low/high pass LCL filter at the home entrance for splitting of the low frequency audio with the above breakpoint high frequency Phone modem line . I estimate that, this should be done within 10m of the home entrance to prevent the spider or cascaded tapped lines from impedance interference dedicated at the home entrance.

(I don’t know what your phone techs do , but this is what I would recommend.)

16Mbps DS with 100k to 500k US requires a very highly processed SNR in each 4.1x kHz sub-band somewhere in the region of 40~50 dB is my guesstimate to achieve N=15bits/baud. (I suppose some keener may compute this or look it up)

The resulting phase distortion determines the ultimate SNR that makes it possible to modulate 8 or more or less bits per sub-channel explained by Shannon-Hartley theorem for probability of error from inter symbol interference.(ISI)

So check with your ISP provider to determine if they can boost the signal levels for the extreme cable length you have, or get a better line.

Your distance from the CO (central office) is the main issue. The home phone line filter location of the hi/Lo filter installation is one possible cause of degradation if not done at the home entrance from long voice stubs in the home.

Conclusion

The stubs of 15m are 1/10th of 2MHz may be tolerated., but if the signal is already marginal 1/20th of a wavelength may be better and no stubs with the Modem splitter filter is best at the home entrance with a distance perhaps of 1mi or less.

CB and HAM transmitter towers near you cable to the CO have also been known to interfere with ADSL2+ phone Modems.

\$\endgroup\$
  • \$\begingroup\$ Unfortunately ISP is very uncooperative in that regard. Since they sell the service as "up to 16 Mbps", they say "your cabling is bad" and that's it. It's not possible to get to the technician who would increase the power. Getting a better line is also not an option since they did not convert the curbside boxes to active fiber nodes in this area (i.e. another line would be no better then the one I have, most likely). For your other comments, I made some updates in my original post. \$\endgroup\$ – alokoko Feb 8 at 19:18
  • 1
    \$\begingroup\$ Cancel service and get cable data \$\endgroup\$ – Sunnyskyguy EE75 Feb 8 at 21:30
  • \$\begingroup\$ I tried that about 5 years ago. It was a painful experience with constant disconnects and all that. The technicians were helpful in that they came and personally inspected the setup and made adjustments to power, but it did not help. One of them finally said the box at the building entrance needed to change since it was from the analog TV era and was not quite suitable for two-way communication. Since they did not do it and did not suggest me a way to have it done, I cancelled the service. The ADSL service is actually rock solid, it just does not perform as good as I would like it to. \$\endgroup\$ – alokoko Feb 10 at 11:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.