Spectrum fragmentation is indeed a challenge. So is preparing the network and making the signaling transition to IP protocols.

SS7, the intelligent signaling layer in TDM networks, is replaced in these LTE and IMS networks by two protocols: SIP and Diameter. SIP is the call control protocol used to establish voice, messaging and multimedia communication sessions. Diameter takes on the role of exchanging policy information between network elements for each subscriber data, voice, video or multimedia session. Diameter messages are exchanged between visited and home networks to authorize roaming subscribers as well.

So service providers and wholesale carriers/roaming hubs, need to build out the network to  manage and secure all this Diameter signaling.

There are a few resources out that defne the problem and requirements, including:

http://www.acmepacket.com/Collateral/Documents/English-US/_8presentations/APKT_LTEroaming_May2011.pdf

http://www.syniverse.com/files/LTE-Roaming-BARG-Workshop-bj.pdf

Yes, the RF spectrum is a major challenge for LTE roaming, but has anyone seen a 3GPP mobile broadband roaming solution between operators? It looks like LTE will still rely on the good old GPRS roaming exchange from R99! That system is already inadequate for a decent HSPA data roaming experience...

PLease check the activities on Multi-Band Multi-Mode at NGMN. http://www.ngmn.org/de/workprogramme/multibandmode.html

This is about assessing technology availability and demand side expectations, both for near and long term, for multimode multiband UEs.

I wonder if it's worthwhile to reinvent the SIM card so it contains the RF module as well.  Using a forwarding service like Google Voice it should be easy to register a local number in Switzerland to my inbound US number (I contact GV using a cert stored on my phone and it registers itself as a specific device).  That would allow for a universal phone and put the cost of RF fragmentation back ot the carriers.  Perhaps difference in price for me the consumer, but at least my US phone has the chance to be used in other countries/networks by buying a local "SIM" without changing my phone number.  

A more interesting side effect of this is that the phone's local number can be more arbitrary, we can go back to the pre-LNP days where the number could deterministically be routed to a specific carrier.  Or perhaps we do away with phone numbers for these devices altogether and just assign them IPv6 addresses, call my number and it can just do a DNS lookup to figure out how to route the call... 

Could be, guessing we're going to have people talking up software-defined radios in glowing terms too...

I wonder if this could be a driving factor to bring tuneable RF modules into commodity pricing over the next few years?  I remember when tuneable optics were for six figure+ test equipment and these days it's often cheaper to only offer tuneables instead of maintaining 40-some ordering codes for wavelength-specific cards.  Not knowing as much about RF design as I'd like I'm uncertain about the difficulties around this, but I know I'd pay a lot more for a global LTE phone than I would for a dual-mode CDMA/GSM phone.

Yep, we could end up in the bad old days of severe NTT DoCoMo LTE-phone envy.

With 40 different spectrum bands for LTE, including both FDD and TDD modes, not only is roaming a problem but so is global economies of scale for devices.  Gone will be the days of producing a single SKU that can be sold worldwide. In a spectrum fragmented world the cost and availability of devices will likely be a major problem.

It doesn't appear that you'll be able to roam between AT&T and Verizon 700MHz LTE networks because they use slightly different parts of the band and would have to put in another RF element in each device to support roaming.