HSDPA Performance in Operational Networks Part 4

My recent trip to Italy produced quite a large amount of measurement data while I was using Telecom Italia Mobile’s (TIM) HSDPA network for everyday work and pleasure. In part 1, I’ve been giving a general overview of the performance of HSDPA in an operational network. Part 2 then focused on analyzing IP packet inter-spacing and revealed a number of interesting details of the HSDPA MAC layer. In part 3 I showed how antenna position and placement can have a tremendous impact on performance. This part now picks up the thread and shows how the HSDPA MAC layer adapts to the antenna position changes tested in part 3.

The picture on the left shows the packet inter-spacing diagram generated from the same data as the throughput graph presented in part 3. If you haven’t read part 2 which gives an introduction of how to read this type of diagram I strongly recommend you to do so before reading on. The throughput graph in part 3 starts off with a speed of around 500 kbit/s. In the packet inter-spacing graph the reason for this slow speed becomes apparent. Instead of most packets being transmitted with a inter-spacing of 10 ms or less, most packets are rather on the 20 ms and 30 ms lines. This either means that the Node-B has sent the packets with a more robust coding scheme or that most packets were retransmitted at least once. No exact telling without a L1 tracer but I guess it’s a more robust coding scheme.

By changing the antenna position the data rate suddenly increases to over 1.500 kbit/s. In the packet inter-spacing graph this is reflected by most packets being transmitted with the least robust coding scheme on the 10 ms. A certain percentage of the packets are retransmitted and show up on the 20 ms line but they are not many, around 20% I would say.

This interval with good signal conditions and the resulting  good transmission speed is then followed by vary bad signal conditions. Here, the data rate drops to around 350 kbit/s. In the packet inter-space graph there are almost no frames transmitted on the 10 ms and 20 ms line. The first major line is at 30 ms. Surprisingly there is not only major additional line at 40 ms but also at 50 ms. Some packets even have an inter-space time of 70 ms. I was quite surprised by this at first. I did some reading in the meantime, however, and saw that Harri Homa and Antti Toskala in their book "HSDPA/HSUPA for UMTS" describe in figure 7.32 that during bad signal conditions their test system did not select only a single but more robust coding like in good signal conditions but diverged between 700 and 1700 bits of user data per 2 ms frame.

There remain those packets during good signal conditions to be explained which are between 2 ms and 10 ms. An interesting point here is that there is not a single IP packet between 0 and 2 ms. A clear indication of the 2 ms MAC layer frame duration. My best guess concerning these packet inter-space times is that they follow a packet which had a transmission error and were transmitted before the faulty MAC layer frame could be retransmitted. This is one of the strengths of the HARQ (Hybrid Automatic Retransmission Requests) used by HSDPA on the MAC layer which continues sending higher layer packets even if a previous one has not yet been transmitted successfully.

And this thought concludes today’s HSDPA entry. Should my German Vodafone prepaid SIM for data roaming have arrived when I come back home more entries will follow soon on HSDPA performance in other countries. So stay tuned…