Televic Rail has a product called MAR - Mobile Access Router, that is actually a wireless gateway put on a train. It connects on-board with off-board, also called the wayside. WiFi b,g,n and 2G,3G,4G (LTE) are supported. Because a train moves fast, there are often handovers, and often the wireless signal is lost.
Now we (Network-Manager daemon on Debian Linux) only react on a changing wireless signal quality, but we are always too late, resulting in a relatively long connection-loss time during handovers.
Network connectivity could be improved by continuously monitoring the wireless signal quality, and by using patterns found in the signal quality coupled with time-of-day and the followed route (hence the journey ID).
Big data collection and analysis
This is the subject of another thesis “Self-learning Mobile Access Router to improve wireless link efficiency”
Adaption of wireless gateway.
The current Network-Manager (NM) version only changes the connection when it fails. But before failure (Access Point (AP) out of range) the data throughput is 0. And when selecting an AP, not the best is taken, but the same as the previous (the smallest time stamp) is taken.
So first step is to (pro-actively) change the connection when a better one becomes available. Maybe this is already present in a new version of NM, or NM must be adapted, or another program must be selected.
Second step is to use the outcome of the self-learning wireless link quality data:
Pro-actively change/connect to the best depending on the use case
Enable to keep 2 links simultaneously: a big file upload prefers free WiFi, a real-time audio announcement maybe prefers reliable 4G
An extension (if time permits) is to use a secure connection. Currently we use a VPN tunnel. Is it possible to make the application (file download, or SIP audio) un-aware of those wireless connection changes, so that it always sees the same VPN tunnel?
To transport high volumes of non-real-time data, we prefer to use the free WiFi, for example a database update. That update must be present on a certain date. To download a big amount of data, a conservative estimate could be that it will take 1 week, so if the data could not be downloaded over WiFi 1 week before the activation date, it must be downloaded over 3G. If one could predict the WiFi availability, the estimate could be much more accurate, saving money.
The current adaptive algorithm of the MAR acts on an improving link by transporting the data in bigger chunks and vice versa, but this always comes too late, wasting costly bandwidth. With the learned signal quality, the bandwidth could be optimally used.
Audio announcements from the wayside (e.g. the control center) must be heard on (selected) trains. The announcement must be understandable (intelligibility), so packet loss must be very small. So it could be beneficial to use 4G even when WiFi is available, because you do not have the location-based WiFi quality and handover times. But with the learned signal quality, the optimal wireless technology could always be used.
Type of work: 30% research/ 60% implementation/ 10% experimentation
Location: School/Televic site/ Field
Number of students: 1