Testing A Connection

13th July 2018 by jude

Our three tests can run for an extended period of time to give good indications as to how voice will perform on your network.

There are almost too many variables to assess whether an internet connection is of sufficient quality to support voice.  

The problem with raw speed tests is that they measure total capacity but you can miss delay or latency.  The theory with speed tests is if the connection is fast enough it can overcome a series of other issues. There is a speed test with the test suite however we have found it to be of limited benefit, if the connection speeds are slow. Even with very fast connections we have found there is a delay that will affect voice. 

We have found three tests that we can run for an extended period of time that give good indications as to how voice will perform.

Jitter Test:

The Jitter test looks at the round trip time of a series of data packets going to the server and back again. This is called the packet latency.  A single round trip (ping time) is too small a test to give a picture of the variability or stability of the connection.  So by measuring a series of ping tests we get a picture of the variability of the connection.  It is a statistical measure of the delay / latency of a series of data packets.

Time Series Jitter Test:

 

This test can be run as a one off picture 35 pings from a chrome browser.  Or you can take a VOXBOX which will plug into your network and allow you to schedule a series of tests over time. You set the frequency of test times.  The maximum frequency of this test is four tests a day, as it uses a full download test which can interfere with networks.

With the VOXBOX tests the results are sent to our servers and then graphed to give you a real sense of how the network is performing.

Call Quality Test:

This test plays a voice file from your network to our servers and back again.  It compares the pre-recorded output file with a call recording of the return file to statistically measure the similarity of the two files.  It uses software similar to the music streaming service Shazam.  You can listen to the return file (what you would hear in a phone call) so that you can also asses what the statistical call quality scores really means.  The software in this test is really interesting.

The following steps are used in the test.

  1. Record the return leg of the call and upload it to our servers.
  2. Re-sample the recorded call at a rate of 12,000 samples a second to get file that matches the format of the original file.
  3. Generate a spectrogram of both voice calls.  The original and the recorded file.  A spectrogram is a visual representation of the sound frequency of the call.
  4. Create digital fingerprints of these spectrograms looking at their intensity peaks.
  5. Overlay these digital voice fingerprints to statistically compare their similarity.

Assess what the score means by listening to a download of the return leg of the call, so that you can see what a real call sounds like over that connection.

Once again the test can be run as a one off from your chrome browser.  Using the VOXBOX apparatus it can be plugged into a network and undertake a series of tests.  It can run tests every fifteen minutes, or on the hour.

Long Call Test:

This test plays a call over a period of an hour and measures in real time the latency, packet loss, and round trip time.  It is very useful if you are diagnosing a network as it allows you to stress the network as the call is going on to see the effect on call quality in real time.  

There are several ways of stressing the network. One way is to schedule a speed test at the same time as the long call test.  This should stress the network as it is assessing how quickly a connection can work.  

The real time round trip delay looks at the round trip time of the packets of data on the call.  The graph shows the average for the previous five seconds. So you should see when the network recovers from some stress test, as the call keeps going.

The lost packet test just adds the lost packets over the hour.  Voice calls can accept up to 1.5% loss packets evenly spread over time without a noticeable deterioration in voice.

The jitter tests gives a real time feed of jitter over the previous five seconds.

These tests are very useful to network engineers as they tune a network.  They can see how it is performing in real time as they make a series of adjustments.

Testing the network rather than the connection.

Testing the network should be done at the entry or egress point for Internet traffic.  Normally this is directly into a router. You just plug the VOXBOX into the router through a ethernet cable.

To test the network just use any RJ45 connection point at the edge of the network.  This would normally be a phone connection socket.  You could plug in multiple VOXBOX into multiple sockets to stress the network.  The difference between the scores that you get between the router tests and the RJ45 tests is an indication of how the network is performing.  Normally the test results should be pretty similar, as a well-designed network should have negligible effects on the performance results.  

If there is a substantial difference with all of the RJ45 tests there could be a cabling or switch issue.  If it is isolated to a single RJ45 socket then it is more likely to be a particular cabling issue.

Bringing your network unto date with a dedicated Fibre connection for voice should provide you with satisfactory results and allow you move to VoIP with a good deal of confidence.

Don't forget to share!