Ronja Twibright Labs

Measuring SNR difference between 5 and 10MHz

INFO

What is SNR difference

The frequency response of the amplifiers on the track is not ideal. Therefore 5 and 10MHz frequencies may be transmitted with different strength. That causes bit patterns with 5MHz content to have different bit error rate (BER) than those with 10MHz content.

If the difference is too high, it means that the signal is being distorted on the way and that the electronics doesn't work properly.

INFO

Requirements

  • A computer with 1 free full duplex network card or 2 free half duplex cards
  • If 2 half duplex AUI cards are used, then Half Duplex Reduction (HDR)
  • Linux where C programs can be compiled.
  • GNU R installed
  • Bertest installed
  • A mirror or long enough extension cables for RX or TX
  • Having the electronics not installed inside optical heads

Network card (NIC) setup

If you are going to use just one NIC, set up full duplex and 10Mbps with autonegotiation disabled. Some hints.

Connect the DUT (Device Under Test)

  • With 1 full duplex card, connect the DUT to that card
  • With 2 network cards and half duplex reduction (HDR), connect HDR into the 2 NICs and DUT into the HDR. Transmitting card is the one with "T" end of the HDR and receiving with "R"
  • With 2 network cards you need 2 tested devices (Twisters or AUI Forte's) and plug one into each card.

Running pktloss

If your input and output network card is eth0, run pktloss. If eth1 is input and output, run pktloss eth1. If eth1 is output and eth2 input, run pktloss eth1 eth2.
INFO

No DUT connected

In such a case the output will be like this:
1102885167.075454000    100.000 100.000 1024    2224
1102885167.645578000    100.000 100.000 1024    2224
1102885168.215306000    100.000 100.000 1024    2224
1102885168.785415000    100.000 100.000 1024    2224
1102885169.355167000    100.000 100.000 1024    2224
1102885169.925561000    100.000 100.000 1024    2224
Second and third column show packetloss at 10MHz and 5MHz in percent.
INFO

Device with packetloss connected

In such a case the output will look like this

:
1102885377.652586000    13.868  77.637  1024    2224
1102885378.222727000    14.649  80.372  1024    2224
1102885378.792451000    17.969  83.692  1024    2224
1102885379.362579000    17.188  85.157  1024    2224
1102885379.932404000    22.364  84.961  1024    2224
1102885380.502458000    15.528  79.883  1024    2224
1102885381.072232000    13.379  76.661  1024    2224
1102885381.642390000    15.918  75.196  1024    2224
1102885382.212091000    12.793  87.012  1024    2224
Second and third column show packetloss at 10MHz and 5MHz in percent.
Drawing
Postscript / PDF / EPS / BIG png / SVG (Inkscape)

Distance adjustment

Run the measurement and adjust the distance between RX and TX so that both second and third columns show numbers with 50 roughly between them (0 and 100 should never occur if possible). Example output:
1102885377.652586000    13.868  77.637  1024    2224
1102885378.222727000    14.649  80.372  1024    2224
1102885378.792451000    17.969  83.692  1024    2224
1102885379.362579000    17.188  85.157  1024    2224
1102885379.932404000    22.364  84.961  1024    2224
1102885380.502458000    15.528  79.883  1024    2224
1102885381.072232000    13.379  76.661  1024    2224
1102885381.642390000    15.918  75.196  1024    2224
1102885382.212091000    12.793  87.012  1024    2224
[...]

Measurement

Choose output filename describing the measurement you are doing, we'll use example.dat for example. Let it run with output redirected into a file this way for a couple of minutes:

bertest <arguments> > rx_1.dat

Running GNU R

Now run ./analyze example.dat (replace example.dat with your datafile name). example.ps and example.pdf will be generated.

SNR difference

Display the resulting Postscript or PDF and goto page 17. Read SNR difference from left edge of graph at page 14.

Evaluation

I am getting values like 0.7dB or 1dB. I consider these normal.

An expected information missing here?