Here's an interesting thought I had the other day I'd like to throw at you to get some feedback. The main challenge faced by LTE is that the current GSM and UMTS voice infrastructure can not be used with LTE anymore as it is an IP only network. There are a number of solutions to the issue such as VoLGA, VoLTE and CS fallback but let's assume just for a minute that neither of them is able to get any traction due to one shortcoming or another. There is one more solution and that is dual-radio, the LTE radio is active at the same time as the GSM or UMTS radio.
Verizon, for example does this already, as shown with their HTC Thunderbolt. The Thunderbolt uses their CDMA 1xRTT network for voice calls and their LTE network for IP based traffic. For the CDMA part, CDMA credentials are used that are stored in a secure place in the device while the LTE security is handled by a SIM card.
So far so good. But what about countries that use GSM/UMTS? Here, a single set of security credentials, or, in other words, subscription is used so one can't be registered to, for example, GSM and LTE at the same time. At least that's what I thought so far. But is it really so? If, let's say, only a GSM circuit switched location update is performed, only the MSC is registered in the HLR/HSS. The packet switched part on the other hand is still open. Now the mobile device registers (only) the packet switched part over the LTE network. This would trigger the current MME to be registered with the HLR/HSS. In other words, there is no collision with the CS part of the network at all of this registration.
If the mobile runs out of LTE coverage it could then perform a routing area update in the UMTS network and also move the circuit switched connectivity from GSM. And once LTE coverage is available again, it could move the packet switched connection back to LTE while the circuit switched connectivity could either move back to GSM or stay on UMTS. From a network point of view I don't see anything that would speed against it. Do you?
Now how easily this could be implemented from a mobile device point of view is another matter. If there's an integrated GSM/UMTS/LTE baseband than that's likely to be difficult. But if there's just a UMTS/LTE chip for IP connectivity and a separate GSM chip for voice connectivity then the implementation would be straight forward, except for the physical access to the SIM card that would have to be multiplexed.
Obviously the power consumption of a dual radio device is higher than that of a single radio device. However, the power consumption of a GSM baseband while in idle is very small. While it would definitely have an impact to the idle standby time which is measured in the hundreds of hours these days, I think the impact it would have on total running time is much less, especially in smartphones. Most smartphone users make good use of their device today and regularly have to recharge it once per day anyway. So a hundred hours less standby time or so won't really matter very much.
Coming back to my question above. Perhaps I'm blind but I can't see any obstacle from a network point of view to have a mobile device CS attached only over the GSM network while PS only attached over UMTS or LTE. The HLR/HSS should not care at all. Let me know if you think I'm wrong.