One of my favorite topics is cellular network capacity. I posted an example a while back on the 1 kb/s 3G surfer. At the time I excluded VoIP as part of my application mix as it increases network traffic quite a bit. Today I found some interesting material which sheds some light on this part of the story:
Wireless Networks have a particular problem with Voice over IP. While traditional circuit switched traffic was optimized on all layers of the protocol stack to be transferred as efficiently over the air interface as possible, achieving the same effect for Voice over IP is very difficult due to the decoupling of the different network layers on the IP protocol stack. Thus, a VoIP call today consumes at least four times as much bandwidth on the air interface than a circuit switched voice call. In other words, if everybody started to use VoIP over wireless today, network capacity for voice calls would shrink to only a quarter of what it is today.
Now Ericsson has released an interesting slidepack on the topic which targets the non geek investment community but which nevertheless contains some interesting numbers on wireless network enhancements and optimization for VoIP in the future: As described above, Ericsson’s slide pack shows on page three that UMTS networks today could only provide 20% of the voice capacity with VoIP compared to standard voice calls. So that’s close to my number above. With HSPA (I think they refer to HSDPA + HSUPA) capacity increases to 70% if IP robust header compression (ROHC) is used. They then go on to claim that with efficient signaling (whatever that is…) and improved scheduler (what improvements?) and optional GRAKE2 (again what is that?) VoIP capacity can be pushed to 140% of today’s standard voice capacity. Finally the next evolution of UMTS called LTE (Long Term Evolution) targets 200% of current voice traffic for VoIP in the same bandwidth.
Not sure what the improvements are they are talking about as they don’t give any further explanations but it seems we are getting somewhere in due time. VoIP VoIP Hurray!
Hi Martin,
Nice VoIP summary.
Just wanted to comment on some of the features that Ericsson is suggesting to improve VoIP/UMTS capacity.
HSPA schedulers:
HSDPA was designed from the onset to deal with peaky types of data transmissions, and the scheduling algorithms are designed to make the most out of the instantaneous good channel conditions for individual users. This is not good for applications that exhibit roughly constant bit rates and are tolerant to delay. Therefore scheduling algorithms that are designed around reduced packet delay are desirable in this case, hence the need for “enhanced schedulers”. Contrary to what Ericsson suggests this means that the VoIP/HSDPA capacity will slightly drop because you are trading off capacity with shorter delay.
On HSUPA, the story is much simpler because of the so called “Non-Scheduled Transmission” feature in HSUPA.
Efficient Signaling:
I think this is referring to two ideas. The first is the idea of carrying UMTS signaling over HSDPA/HSUPA which is expected to reduce the additional resources often attributed to dedicated signaling channels as well as reduce the signaling delay. The other idea is “re-structuring” some of the signaling channels to improve capacity.
There are additional methods to push the VoIP/HSPA capacity up, such as diversity. Receive diversity on the terminal side is expected to at least double the capacity, but it creates lots of challenges to the terminal chip makers. Transmit diversity is also possible but it has negative implications on other data/HSPA applications. Diversity is beneficial to traditional circuit voice too, so the delta between VoIP and circuit voice does not change in this case, which probably explains why Ericsson did not include it in the slide pack.