Great re the higher speed, particularly at 800m. I was getting around 70Mbps sync on 3dB, and mine was a pretty clean line.

Re your "understanding", you start right but then get confused. It hadn't switched to 6dB. The noise had fallen so the margin rose. If the noise creator had just turned their modem (or modem/router) off and gone on holiday that would cause it, but when they came back and turned it on again it would revert to roughly what you had before.

On FTTC the sync-time margin is set by DLM and applied at the DSLAM. Your modem basically accepts that then they negotiate the speed that can be maintained. The noise varies continuously however, so the margin your modem reports will fluctuate in the opposite direction to the noise change. The Attainable is just the modem telling you that "Hey - we could go faster, (or slower), than we are doing because the noise level has changed".

The theory is that you lose connection when the margin hits zero, and get a re-sync. In practice it depends on the quality and condition of the actual hardware. Some can lose it at 1dB, others can stay connected on -1dB or even -2dB.

Error rates are likely to be high however when close to losing connection, and correcting those or retransmitting the packets can reduce the actual throughput to well below what you would get with a re-sync at a lower speed.

The base point is only ever set at the time the connection/sync is made. (On Openreach FTTC and BT Wholesale ADSLx). The whole point of the margin, as explained in the page I linked to, is to stop you losing connection when the noise goes up.

ADSL is a lot lower frequencies than VDSL2 so yes dropping to ADSL would remove a crosstalk source

Minor point: the SNR margin isn't how far above the noise floor the signal is, it's how far above the level needed to provide that data rate the signal is.

A tone set to deliver 4096 QAM, 12 bits per symbol, needs better SNR than a tone delivering 256 QAM, 8 bits per symbol. Margin is the level above the needed SNR to deliver almost zero bit error rate for the negotiated modulation order the actual SNR is at.

EDIT: As a rough guideline 3dB more SNR = 1 more bit per symbol. 64 QAM / 6 bits per symbol needs about 28dB, 128 QAM / 7 bits per symbol 31dB, etc. These numbers are without coding gain, which means error correction, so FEC can and does make a difference, as does G.inp.

Edited by deleted (Sun 19-May-19 23:44:05)

SNR margin is the ratio between the current signal level, and the signal level at which it would just become impossible to distinguish signal from noise.

Thus, the higher the SNRM, the clearer the line is; this is made a bit more complicated as other posters have said because higher sync speeds mean that the signal is harder to distinguish from noise.

So, assuming you hold sync speed constant, higher SNRM is better - a line with a 9dB SNRM and a sync of 80,000 kbps is "clearer" than a line with a 3dB SNRM and same sync speed. If you hold SNRM constant, then clearer lines have higher sync speeds - a line that gets 80,000 kbps sync at 3dB SNRM is clearer than a line that gets 68,000 kbps sync at 3dB SNRM.

In reply to a post by farnz:

SNR margin is the ratio between the current signal level, and the signal level at which it would just become impossible to distinguish signal from noise.

Thus, the higher the SNRM, the clearer the line is; this is made a bit more complicated as other posters have said because higher sync speeds mean that the signal is harder to distinguish from noise.

So, assuming you hold sync speed constant, higher SNRM is better - a line with a 9dB SNRM and a sync of 80,000 kbps is "clearer" than a line with a 3dB SNRM and same sync speed. If you hold SNRM constant, then clearer lines have higher sync speeds -

Only for any particular line and attenuation. Attenuation is the prime variable in most cases. What you say does not apply in the general case.

For instance, "a line that gets 80,000 kbps sync at 3dB SNRM is clearer than a line that gets 68,000 kbps sync at 3dB SNRM" is not a valid statement. Unless more information is available the assumption would be the 80Mbps one is shorter.

Edit: SNRM is not a ratio anyway. SNR is. SNRM is the arithmetic difference between the current SNR and the SNR at which connection would probably be lost. A little difficult to explain simply, given the fact that SNR is a logarithmic scale.

Edited by RobertoS (Mon 20-May-19 16:10:30)