Converged networks, blended applications, a revolutionary re-creation of the wireless services architecture — there's a lot of fancy rhetoric being tossed around when the industry talks about IMS. But one of the major elements to IP Multimedia Subsystem migration, VoIP, seems to get lost in the cacophony.

While IMS elements will go into the network core long before VoIP will make its way to the radio access network, a lot of IMS' key features depend on end-to-end VoIP. The whiz-bang features that are often touted during IMS presentations, such as push-video applications, multimedia calling and presence-based voice services, require tossing voice into the bit stream along with other IP applications. Converging wireline and wireless numbers requires the session initiation protocol (SIP) stack that powers today's VoIP phones but as yet is absent in the handset. And even the benefits of the IMS architecture, which is supposed to make the creation of new applications a breeze for the carrier, will face hindrances if it has to shoe-horn circuit-switched voice into its neat, clean SIP-based services creation layer.

“Simply put, you don't get that seamless transparency between applications without VoIP,” said John Marinho, vice president of strategic marketing for Lucent Technologies.

The coming of VoIP to the wireless network is therefore an inevitable event, and many carriers have already committed to deploying it. Sprint has said it will start testing EV-DO revision A late next year; it will begin radio access network (RAN) trials of VoIP over the new broadband network soon afterward. The goal is to have VoIP powering push-to-talk (PTT) in 2008. NTT DoCoMo in Japan, meanwhile, has said it doesn't plan to have a spec of TDM left in its network by 2012, just an all-IP network from handset to handset.

Even if those dates seem far off, they're nothing compared to what most industry experts predict will be the timeline for the full implementation of wireless VoIP. Carriers seem committed to getting the most out of their 2G voice networks for at least the next 10 years, and in an industry driven by long standards processes, the discussions of how VoIP will be implemented in the RAN is just now beginning. VoIP is no doubt coming, but it will take a while to happen. In the interim, the industry is figuring out just how it will arrive, where in the network it will be implemented and finally — and perhaps most significantly — what exact role VoIP will play in the IP networks of the future.

It's possible to make a VoIP call over the cellular network today. There are clients such as Skype's customers that can load onto a wireless PDA and use the IP data connection to access a VoIP carrier's network. The performance of these solutions leaves something to be desired, though. The UMTS and EV-DO networks out there today have not been optimized for VoIP traffic, to say nothing of the GPRS/EDGE and CDMA 1X networks. Since the services come from a third party, the voice call is dumped into the bit stream with all other data traffic assigned no priority. A carrier VoIP network would have quality of service (QOS) controls built far out in the network, at least to the base station and possibly to the handset, as well as more sophisticated packet routing capabilities interred in its packet data service nodes (PDSNs) and home agent gateways. Also, the solutions are software-based, depending on the stability, resources and whims of the operating system, instead of embedded directly into the chipset modem. VoIP over the network today is a possibility, but it's a meager possibility at best.

“You don't want one network optimized for voice and one network optimized for data, but you can't just have a packet-based network to run VoIP,” said Jon Hambidge, vice president of marketing for IP access infrastructure-maker IPWireless. “You still need to create a virtual network within the packet network, one that layers data underneath voice and gives priority to multiple types of applications like gaming and video. You can't just do that in the network. You do it at the radio interface.”

The first step toward bringing IP to the wireless last mile is not through the RAN, but through the wireless LAN. While the two networks are completely separate, the drive toward converged devices will not only allow carriers to dump pricier cellular traffic onto Wi-Fi networks, but it will also become a critical factor in pushing IP voice into the radio access network.

While Motorola is promising its first Wi-Fi/GSM phone this summer, most of the industry doesn't expect converged handsets to be available until 2006, when the first unlicensed mobile access (UMA) compliant devices come out. Those devices, however, won't be true VoIP devices. They'll have the IP stack necessary to deliver voice over the wireless LAN, but instead of making an SIP call like a VoIP phone, the GSM call will merely tunnel through the IP connection. That means the full range of IP enhanced services that one normally associates with VoIP won't be available in the first UMA devices, because the Wi-Fi network will still separate voice from data, just like the cellular network.

“There's a distinction here,” said Doug Makishima, marketing vice president for Intoto, an IP services and security software developer. “You're putting IP on the handset, but at the end of the day the phone's still talking GSM to the base station. That IP stack though, will be very important — it will act as a catalyst. Once a phone is IP, all you need is an SIP stack to have a VoIP handset.”

But the end goal of UMA's supporters is to deploy a full SIP stack on the phone. When that happens, the handset will essentially become a VoIP-enabled terminal. From there, it's only a matter of a software upgrade — the same SIP stack that powers a VoIP Wi-Fi call can be used to power a VoIP cellular call.

Perhaps the biggest obstacle to overcome when introducing a new technology is handset replacement. Research firms predict that by 2010, 55 to 100 million dual-mode cellular/Wi-Fi devices will be in the market. If those terminals have full SIP stacks in them, half the battle is already won. Carriers could feasibly have VoIP-ready cellular handsets in their networks before they've deployed VoIP-supported radio infrastructure.

While there is no question that carriers want VoIP in their networks, there is still a big question as to where they want it and how much of it they want. VoIP has already penetrated into carriers' transport networks, and it's quickly making its way into the core as carriers layer in IMS and softswitching. But the RAN is a different animal altogether. Carriers worldwide have invested untold hundreds of billions of dollars into their 2G cellular networks. They're unlikely to be willing to give up that investment for a while — if ever.

Current 2G and 2.5G bandwidths are too narrow to ever support voice, and those networks are already optimized for circuit-switched voice — far more optimized than they could ever be for IP Telephony. While upgraded UMTS and EV-DO networks will soon have the capacity and latency to support VoIP, carriers' first inclinations will be to deploy high-revenue data applications over those networks, not rapidly commoditizing voice services, said Steve Shaw, director of marketing for Kineto, an IP convergence technology developer.

“I don't think there is a need to migrate voice entirely to IP,” Shaw said. “If it's just a question of price, what's the point? If a carrier is getting 5 cents a minute for GSM and voice and 4 cents a minute for IP voice, is he really going to want to move that quickly to IP?”

There's even a question of how much capacity and cost savings VoIP will provide. VoIP's biggest proponents concede that an apples-to-apples comparison of a GSM or CDMA voice channel to an IP voice channel of the same size yields only a doubling of voice capacity. But some vendors say that even that multiple of two might be misleading when looking at the overall number of calls the IP RAN can carry simultaneously. Fixed VoIP networks are using codecs of 60 kb/s. If carriers try to maintain that same quality — which they may have to in order to meet enterprise demands for higher voice quality — they won't achieve any capacity savings on the networks, IP Wireless' Hambidge said. They could feasibly run much slimmer codecs as small as 12 kb/s, allowing them to cram five calls into the equivalent of one voice channel. But unless a huge leap in compression technology is around the corner, the voice quality will suffer greatly, Hambidge said.

Capacity, however, is only the smallest reason to transition to VoIP, said Lucent's Marinho, and if capacity does become a major goal of the carriers there are technology innovations that will greatly boost it such as multiple input/multiple output (MIMO) technologies, which will add multiple antennas to the handset. The major benefit of VoIP in the RAN, though, is its ability to merge seamlessly with the application stack, Marinho said, instead of being the isolated application that it is today.

IMS has the ability to do some amazing things with TDM voice that carriers can only imagine today. Even if the voice network isn't end-to-end IP, all things are equal in the core once a carrier makes the full transition to an IMS service architecture. Just like a wireline VoIP provider can use SIP to bridge the circuit-switched and IP networks with features like simultaneous ring and find-me-follow me applications, wireless can implement the same services with an IMS-enhanced core. Carriers can even launch some of the fancier blended services such as presence-based dialing, through which a customer checks a buddy list to see if a colleague or friend is available and initiates a call or text message using that list.

The problem in these scenarios is that all the intelligence of the voice network is in the core, not on the edge, and certainly not on the handset. As soon as an application blending voice and data features hits the RAN, it splits into two parts: an intelligent IP stream accessing the advanced data capabilities or the handset and a dumb voice pipe that thinks it's making an ordinary phone call.

There are limits to how much mileage a carrier can get from this architecture, said Lucent's Marinho. In fact, as IMS services advance and voice becomes more intertwined with other features of the phone, Marinho said, it will become increasingly difficult — and costly — for carriers to shoe-horn TDM voice into the mix. And instead of the ease of mixing and matching applications on the fly — IMS' most eagerly-anticipated feature — carriers will find themselves reverting to the old model of purpose building individual services around a cantankerous and inflexible voice application.

“What you're doing is mixing IP and circuit-switched applications — which is certainly possible,” Marinho said. “What you don't get, however, is that seamless transparency between applications. Only by merging the two does that seamless network become more readily realizable.”

For that reason, almost everyone agrees that VoIP will make its way into the RAN in some fashion. But VoIP's role may be much downplayed. Despite many carriers' claims that they envision a complete transformation from TDM to VoIP, they're unlikely to flip the off switch on their 2G networks for some time. Instead, carriers may opt to distinguish between voice calls and enhanced calling on their networks. Just a simple person-to-person phone call will probably run on the typical GSM or CDMA 1X network for some time in the future, but when a customer wants to add video, presence or some other data feature to that voice functionality, the call will turn into a VoIP one running over the EV-DO or the UMTS network. And any new service that has a major voice element, such as push-to-talk, will also traverse the network as VoIP.

Carriers will essentially run two different voice networks: a next-generation network for high-revenue blended services and a plain circuit-switched voice network, Kineto's Shaw said. From an infrastructure standpoint, this setup makes sense. To go entirely IP, a carrier would have to upgrade its entire access network to either EV-DO revision A (in the case of CDMA) or high-speed uplink packet access, or scrap wideband CDMA entirely and move to time division-CDMA (in the case of GSM carriers). That conversion would be daunting for a CDMA carrier but unthinkable for a GSM carrier.

Even in areas where 3G networks are built out, carriers may be hesitant to dump voice onto their IP networks. It's just simple pragmatism, Shaw said: If the majority of carriers' traffic is still the basic voice call and they have in place a huge 2G infrastructure optimized to support those calls, why not keep using it?