VoIP Episode 8: The future of mobile VoIP

I'm linking to an article by Malik Kamal-Saadi on telecoms.com because it has a very good treatment of the subject and in essence summarises most of the VoIP/cellular strategic issues I've touched upon in my previous posts. (I don't know why they did not acknowledge the author's contribution on the telecoms.com site. I've received the written report which clearly shows his name).

One of the interesting points that Malik talks about is the high performance requirements of VoIP cellular handsets. Incidentally, both the new 3X series phones offered by Hutchinson 3 in UK (Nokia N73 and Sony Ericsson W950i walkman) can run a Skype client. Nevertheless there is no doubt that having to support HSDPA/HSUPA with all the R6 and R7 enhancements together with enough processing power to support VoIP calls will definitely be a challenge for UE chip makers and vendors alike.

The threat of VoIP to traditional mobile voice revenue may be an inevitable outcome. In my opinion, VoIP's threat is starting to materialise when competing against the roaming chunk of the voice revenue, where business people for instance find cheap or free WiFi networks to camp on and make international VoIP calls. Apart from that, it does not seem that VoIP is giving operators much worry. Three for example has kept the traditional (line rental) pricing schemes, and introduced the X-series broadband package which the user has to pay for extra. The package already includes a large number of free minutes and TXT allowance. It does seem to me that Three has found a clever way to push their ARPU up without cannibalising their existing revenue, at least in the short term.

One of the barriers of VoIP over cellular is its lack of seamlessness. Not any user will appreciate that to run a VoIP call you have to start a different application than the usual button pressing you do when you call someone the usual way. The fact that the voice call is carried over IP should be transparent to the end user. Handset user interface software has to integrate both experiences in one unified package to give the end user the seamless feel. To make an analogy, just like the misconception that mobile Email is simply a pop3 client on your handset disregards usability issues and user experience, so does the flawed idea that mobile VoIP is simply a Skype or Vonage client.


© Copyright belongs to the author. Do not cite or quote without the express permission of the author.

VoIP Episode 4: VoIP over UMTS high speed channels

In the first episode we discussed how the various implementation options compare in terms of performance and complexity. We say that there are various necessary changes to the high speed channels to make them an attractive medium to carry VoIP traffic. Some of these changes are also necessary to reach VoIP capacities that exceed conventional circuit switched capacity. In this post I will discuss some of the suggested changes to the high speed channels to make them suitable to carry VoIP.

• Scheduler changes:
  3GPP defined the structures of the high speed shared channels such that they are optimal to carry large chunks of data. To make the system reactive to users' channel conditions, an intelligent scheduling entity that usually resides in the RNC is moved to base station. Thus if a user happens to be in good channel conditions, the scheduler allocates resources and send them to the user before the favorable channel conditions disappear. The problem with VoIP traffic is that the packets have to be delivered to the user pretty frequently in order to maintain the perception of a good quality connection. The packets can not be queued forever waiting until the user has good channel conditions. Therefore the scheduler algorithms, which were originally developed to be optimal to carry discontinuous chunks of data, have to be adapted to work well with VoIP traffic side-by-side. One of the concepts to enable this is called “priority queuing" which as the name implies, gives priority to delay intolerant traffic such as VoIP. Most of what I describe above applies to HSDPA (downlink). In the uplink, 3GPP included enough functionality in the HSUPA requirements to make it versatile and able to cope with VoIP traffic using a scheme called “Non-Scheduled Transmission”.
• HARQ Retransmission Reduction:
  The HSDPA channel uses also a concept called "HARQ retransmission", where lost packets are retransmitted (partially or completely). The idea is to optimize the air interface efficiency, so instead of sending few highly protected bytes that will be definitely received correctly, the transmitter sends a big chunk of bytes and hope for the best. If part of the data is received incorrectly, then the transmitter can retransmit the packet. This approach is one of the enablers of high speeds achieved on this type of shared channels (see note 1 below). It turns out that the HARQ retransmission concept is a challenge to smooth VoIP operation. This is because the VoIP connection is intolerant to delay, and obviously packet retransmission will increase the round-trip delay, which degrades the call quality. Packet retransmissions also increase the severity of packet jitter at the receiving end: one packet may arrive almost immediately and the next packet may be transmitted several times before successful reception happens. If you eliminate re-transmission altogether the channel efficiency drops and you end up with a reduced cell capacity. (Check note 1 below) .
• Reduce the possibility of code limitation (Fractional DPCH):
  Every user connected to the network using a shared channel requires a number of small dedicated channels to carry signaling and control information. These channels carry very little traffic, but they are numerous. If there are many users in a cell using the HSDPA channel, a large number of dedicated channel is required to support them, and it could happen that the system runs out of channelization codes (if you don’t understand the concept of channelization codes, check this). Therefore, 3GPP included a new feature called Fractional DPCH, which essentially multiplexes the control information of multiple users on one physical channel.
• Continuous Packet Connectivity (CPC):
  This is a huge 3GPP work item which includes many aspects that, although not necessary for VoIP to work, but will benefit its operation and increase the cell VoIP capacity. The most noteworthy of these aspects are:
  
  1. DPCCH gating: The control channels have a discontinuous pattern, which reduces the amount of interference and increases battery life.
  
  2. Modified HS-SCCH operation. HS-SCCH is a channel that carries HSDPA signaling in the downlink. This channel typically consumes large amount of power which eats up a part of the cell capacity. Therefore 3GPP came with various suggestions to reduce its effects by restricting it, or eliminate it altogether.
  
  3. CQI reporting reduction: CQI stands for Channel Quality Indicator, which is a message sent by the terminal to the bas station reporting the channel quality so that the base station can decide what to send to the terminal. As we discussed in the scheduler section above, successful VoIP operation is more concerned with the continuity of the traffic instead of mapping transmission to channel conditions, therefore reducing the rate of CQI signaling will reduce overheads without affecting VoIP quality. This means that more bandwidth is available for VoIP packets instead of overhead.
  
  All in all, CPC is expected to increase VoIP cell capacity by anything from 15% to 40%.
  

Note 1: There is an optimal packet loss rate that will cause the system to operate at the maximum possible capacity. On one hand, high packet loss rates mean that many packets are being retransmitted many times and hence the air interface resource is being wasted on these retransmissions. On the other hand, low packet loss rates mean that the transmitter is not taking chances and is protecting data more than necessary. This will lead to a lower cell capacity.

© Copyright belongs to the author. Do not cite or quote without the express permission of the author

VoIP Episode 1: VoIP over UMTS

There has been talk in the market about introducing VoIP over UMTS networks. In this posting I investigate the viability of various VoIP over UMTS solutions. The three options presented here are: VoIP gateway, end-to-end VoIP over shared channels and end-to-end VoIP over dedicated channels.

Clarification of terms:
Before starting, let me clarify the term "VoIP over UMTS". I chose it because the prevalent term "VoIP over 3G" tends to be more generic and encompasses all technologies people consider as "3G" e.g. EV-DO. In this case I concentrate on UMTS only. Some people, when they refer to UMTS, they mean the 3GPP release 4 functionalities, and tend to refer to 3GPP R5 as HSDPA system and R6 as the HSUPA system. When I say UMTS here, I mean all the above. I should mention here that all the discussion that follows is primarily from a radio network point of view.


What drives service providers to think of the possibility of carrying VoIP over a UMTS radio network? Some of the reasons that I hear from time to time:

• Operators think that VoIP can help them utilise their transmission network more efficiently.
• They think that VoIP can help them squeeze a bit more out of their radio network.
• They freak out when competing providers using other access technologies (WiMax, EV-DO ..etc) are offering VoIP. UMTS providers feel the urge to support VoIP to compete.
• Convergence with their IMS networks.
  

Putting all the commercial drivers aside (competition, prices, availability ...etc), the four technical factors that influences the VoIP implementation are:

1. Capacity: Can the system carry a reasonable and viable amount of VoIP calls?
2. Complexity: How complex is the proposed solution?
3. Call Quality: How does the VoIP call quality compare with users' experience?
4. Seamless convergence: How easily can the implementation offer introduction of converged multimedia services and how readily is it compatible with existing and future network entities?
   

So what are the various ways to introduce VoIP/UMTS? There are three methods I can think of, which score differently in light of the four factors discussed above:

1. Using a gateway solution:
   In this case, the VoIP traffic is terminated in the core network, and the voice is translated to circuit switched format to be carried over the radio network.
   UMTS was built from the onset to carry Circuit switched voice and packet switched data simultaneously, so by using the the UMTS circuit switched capability to carry voice traffic the call setup and release times are short. The call quality is also more predictable, sustainable, and is maintained at a reasonably high quality. The disadvantage is that the end-to-end VoIP capability is lost, and with it the ability to provide added services such as Contact Lists and Contact Presence Information. (It maybe still possible to offer the latter types of services on a separate bearer with an order of magnitude increase in network complexity.)
   Another advantage is that all the IP overheads are stripped off at the gateway, so there is less overhead to be carried over the radio, which is often considered the most expensive resource.
   
2. Providing end-to-end VoIP over high speed channels:
   In this case, the VoIP call is carried all the way to the user terminal over high speed shared channels in the uplink and downlink (dubbed HSDPA and HSUPA in the 3GPP standard). The motivation is to use the channels originally envisaged for data applications to carry VoIP packets.
   HSDPA was originally conceived to increase air interface throughput by giving more channel resources to users who experience good channel conditions. Unfortunately, in VoIP case, this leads to degraded radio bandwidth utilisation, because users' channel condition are irrelevant. Instead, link continuity and sustained voice quality are more important. Therefore, new scheduling schemes have to be developed to suit VoIP type of traffic, which unfortunately defeats the whole purpose of a shared channel that is allocated to users based on their channel conditions. In a nutshell, the capacity performance of this option does not look very appealing. Another disadvantage is the increased call setup delay and packet round-trip delay. Essentially users' packets have to stand in a queue to get a portion of the shared channel resources. Luckily, this effect is less of an issue in the uplink (HSUPA) because of a standard feature called (Non-Scheduled transmission).
   
   Nevertheless, there are various proposed schemes to improve the shared channels in order to carry more VoIP calls. For example, Robust Header Compression (RoHC) reduces the amount of overhead in the IP packets. Another idea is carrying the call signaling over shared channel too, which reduces call delays. There are also other improvements to the uplink and downlink shared channels that will increase the capacity up by anything from 20% to 50%.
   
   The proponents of this option argue that the E2E attributes of the VoIP call are maintained. They also argue that on the long term the VoIP capacity is much enhanced by introduction of advanced receivers and diversity (transmit div, receive div or both).
   
3. Providing end-to-end VoIP over dedicated channels:
   In this case, the VoIP traffic is carried over dedicated channels. This sounds a more plausible implementation than using high speed channels: The high speed channel model tries to blast high data rates to users in good conditions, which is great for data hungry applications, but is not advantageous to VoIP users who are more interested in quality than high data rate. The dedicated channels on the other hand adjust the power allocated to each channel (power control) , so each user gets just the right share of the radio resource (give or take). There is still the issue of packet overhead, therefore the capacity performance of this option is better than VoIP/high speed channels, but is still less than pure circuit switched voice.
   The E2E benefits are there too, so it is easy to introduce IM types of services.

So what is the verdict? VoIP over dedicated channels seems to offer the best trade-off between capacity and complexity. It also provides for reasonable VoIP call quality while maintaing enough attributes to enable easy multimedia convergence. On the long run, the introduction of enhanced features may cause the VoIP/HS implementation to perform better in terms of capacity and call quality.

I will discuss the various enhancements in a another post soon!.

© Copyright belongs to the author. Do not cite or quote without the express permission of the author