3 ms·

I confirm the two to four seconds delay in experimental deployments. When I first saw the proposed architecture, I couldn't believe it - kludge was the first wo

by liotier 14y ago

I confirm the two to four seconds delay in experimental deployments. When I first saw the proposed architecture, I couldn't believe it - kludge was the first word to spring to my mind. All that while the deployment of IMS platforms is gathering steam... Why can't IMS become the voice carrier ? SIP everywhere and no more circuit switching... What's wrong with that ? NIH ?

seiji 14y ago
Another "why can't they do this already" technology that impacts millions of people every day: seamless GSM to wifi (and back) handoff. They had specs, prototypes, and demos back in 2004. It never went anywhere. Is it just because "they" wanted to overcharge for microcell hardware and tack on monthly fees? Imagine being on a call outside, walking inside, through an elevator, and up to your room/office/flat/apartment and not have the call disconnect. No GSM repeaters in the loop. Your call went from AT&T->Wifi VoIP automatically and it'll switch back to GSM when you leave wireless range again.
- liotier 14y ago
  Handover between GSM/UMTS and Wifi works in the wild... But there is more to market acceptance than just having a perfectly fine technical solution : http://www.kineto.com/wi-ficellular-handover-the-state-of-the-market/ http://www.kineto.com/wi-ficellular-handover-the-state-of-th...
dfox 14y ago
"VoLTE" seems to be an new term for some subset of IMS. There are three problems with end-to-end SIP as of now: 1) You want to make and receive calls where other end is in PSTN or non-AIN PLMN or otherwise unable to directly process RTP payload stream (you essentially can use SIP for circuit switched data paths, although terminating SIP sessions at adge of circuit switched network seems like better solution). 2) Handover LTE->GSM when leaving coverage area. It is possible to handover an active packet data session from LTE radio channel to almost anything else (including WiFi), but with active call the target transport has to be able to support required bitrate and latency, which is clearly impossible with plain GPRS, barely attainable with EDGE. On the other hand LTE allows transparent handover to just about anything that is IP (wired ethernet, anyone?). Cludgeness of the IP handover architecture is caused by two things: no real deployment of IPv6 when first GPRS networks were designed and desire for this architecture to support non-IP protocols (things like X.25 over EDGE are specified as possible, although probably non-existent). 3) QoS on backhaul network. While LTE All-IP backhauls are simpler to build (and in fact mostly cheaper, because designer has more freedom with network topology), they present different set of challenges than normal circuit switched networks when it comes to QoS. This tends to be complicated further by desire to tunnel PCM/ATM based interfaces of legacy network base station components of newly built sites through same backhaul Ethernet/IP link (there goes your topology freedom...). Also there is slight problem of how to bill SIP/RTP calls spanning different networks, it is technically possible to capture data you need to do normal per-minute billing, but it is not clear whether that is good model for LTE. In my opinion best solution is to bill for data transferred across LTE network, for call-management operations (ie. per-call fee) and for possible costs of call termination through circuit switched network all at once, with possibly separate entities providing each of these services. This seems to me like fair pricing scheme, but on the other hand it is not understandable by common consumer and thus probably infeasible for carriers. Not doing voice calls over LTE and relying on fallback to whatever legacy circuit switched technology is supported by network and MS is good way for carrier to workaround these issues. But if I correctly understand conditions of LTE spectrum allocations in Czech Republic, best way to meet these conditions is to actually roll-out VoLTE from the start (these conditions mostly require availability of voice service everywhere with expectation that missing coverage areas of GSM/UMTS will be filled with LTE, which seems like technological nonsense to me, but it is how it is).
- liotier 14y ago
  Billing has seen much worse cases. QOS is just fine too, at least on the radio segment : LTE allows dedicated bearers, each with differentiated QOS, and no one in his right mind will try VOIP on GPRS or EDGE. Backhaul QOS is currently contentious - which is why we are scrambling like mad to lay fiber everywhere, so it will only be a problem for a while. Inter-protocol handover is hairy, but it has been done before - though it is indeed a significant technical hurdle. So handing over between SIP voice and CS voice seems to be the harder problem. Has anyone seen research in that direction ?
  - dfox 14y ago
    Nokia E-series handsets can be persuaded to support VoIP over EDGE and it sort of works (few years ago we did some testing of that with intention to use that as part of cab telematics), one would say that it works better (certainly it's more reliable) than VoIP over WiFi with same handsets. Interprotocol handoff between two CS protocols is mostly equivalent problem to normal GSM handoff, handoff of packet switched transport is mostly transparent to higher level applications (ie. voice). Clear problem with IP->CS handoff is sudden change of QoS parameters of channel which I think cannot be overcame in any sane way and any solution involving IP->CS handofs will reintroduce audio artifacts of NMT. (hypothetical insane way: pump full RTP datagrams through UMTS radio channel instead of bare audio, which is impossible to do with GSM radio layer). CS->IP handoff (which seems to me to be more interesting in czech environment) has same problems and also lacks any support in existing CS protocols that are mostly set in stone. On the other hand I think that LTE deployment will go reasonably fast so these issues can be mostly ignored by doing fallback to CS where possible and using only LTE when deployment is sufficient. So I don't think that anyone is going to seriously consider solving problems inherent in CS<->IP handoff.

by liotier 14y ago

seiji 14y ago

Another "why can't they do this already" technology that impacts millions of people every day: seamless GSM to wifi (and back) handoff. They had specs, prototypes, and demos back in 2004. It never went anywhere. Is it just because "they" wanted to overcharge for microcell hardware and tack on monthly fees? Imagine being on a call outside, walking inside, through an elevator, and up to your room/office/flat/apartment and not have the call disconnect. No GSM repeaters in the loop. Your call went from AT&T->Wifi VoIP automatically and it'll switch back to GSM when you leave wireless range again.

liotier 14y ago
Handover between GSM/UMTS and Wifi works in the wild... But there is more to market acceptance than just having a perfectly fine technical solution : http://www.kineto.com/wi-ficellular-handover-the-state-of-the-market/ http://www.kineto.com/wi-ficellular-handover-the-state-of-th...

dfox 14y ago

"VoLTE" seems to be an new term for some subset of IMS. There are three problems with end-to-end SIP as of now: 1) You want to make and receive calls where other end is in PSTN or non-AIN PLMN or otherwise unable to directly process RTP payload stream (you essentially can use SIP for circuit switched data paths, although terminating SIP sessions at adge of circuit switched network seems like better solution). 2) Handover LTE->GSM when leaving coverage area. It is possible to handover an active packet data session from LTE radio channel to almost anything else (including WiFi), but with active call the target transport has to be able to support required bitrate and latency, which is clearly impossible with plain GPRS, barely attainable with EDGE. On the other hand LTE allows transparent handover to just about anything that is IP (wired ethernet, anyone?). Cludgeness of the IP handover architecture is caused by two things: no real deployment of IPv6 when first GPRS networks were designed and desire for this architecture to support non-IP protocols (things like X.25 over EDGE are specified as possible, although probably non-existent). 3) QoS on backhaul network. While LTE All-IP backhauls are simpler to build (and in fact mostly cheaper, because designer has more freedom with network topology), they present different set of challenges than normal circuit switched networks when it comes to QoS. This tends to be complicated further by desire to tunnel PCM/ATM based interfaces of legacy network base station components of newly built sites through same backhaul Ethernet/IP link (there goes your topology freedom...). Also there is slight problem of how to bill SIP/RTP calls spanning different networks, it is technically possible to capture data you need to do normal per-minute billing, but it is not clear whether that is good model for LTE. In my opinion best solution is to bill for data transferred across LTE network, for call-management operations (ie. per-call fee) and for possible costs of call termination through circuit switched network all at once, with possibly separate entities providing each of these services. This seems to me like fair pricing scheme, but on the other hand it is not understandable by common consumer and thus probably infeasible for carriers. Not doing voice calls over LTE and relying on fallback to whatever legacy circuit switched technology is supported by network and MS is good way for carrier to workaround these issues. But if I correctly understand conditions of LTE spectrum allocations in Czech Republic, best way to meet these conditions is to actually roll-out VoLTE from the start (these conditions mostly require availability of voice service everywhere with expectation that missing coverage areas of GSM/UMTS will be filled with LTE, which seems like technological nonsense to me, but it is how it is).

liotier 14y ago
Billing has seen much worse cases. QOS is just fine too, at least on the radio segment : LTE allows dedicated bearers, each with differentiated QOS, and no one in his right mind will try VOIP on GPRS or EDGE. Backhaul QOS is currently contentious - which is why we are scrambling like mad to lay fiber everywhere, so it will only be a problem for a while. Inter-protocol handover is hairy, but it has been done before - though it is indeed a significant technical hurdle. So handing over between SIP voice and CS voice seems to be the harder problem. Has anyone seen research in that direction ?
- dfox 14y ago
  Nokia E-series handsets can be persuaded to support VoIP over EDGE and it sort of works (few years ago we did some testing of that with intention to use that as part of cab telematics), one would say that it works better (certainly it's more reliable) than VoIP over WiFi with same handsets. Interprotocol handoff between two CS protocols is mostly equivalent problem to normal GSM handoff, handoff of packet switched transport is mostly transparent to higher level applications (ie. voice). Clear problem with IP->CS handoff is sudden change of QoS parameters of channel which I think cannot be overcame in any sane way and any solution involving IP->CS handofs will reintroduce audio artifacts of NMT. (hypothetical insane way: pump full RTP datagrams through UMTS radio channel instead of bare audio, which is impossible to do with GSM radio layer). CS->IP handoff (which seems to me to be more interesting in czech environment) has same problems and also lacks any support in existing CS protocols that are mostly set in stone. On the other hand I think that LTE deployment will go reasonably fast so these issues can be mostly ignored by doing fallback to CS where possible and using only LTE when deployment is sufficient. So I don't think that anyone is going to seriously consider solving problems inherent in CS<->IP handoff.