08 December 2011

UMTS: Truly, you have a dizzying intellect

We have spent a lot of time in recent months studying the specifications for "Uu", the UMTS 3G subscriber interface. The spec is an obfuscated mess and the experts who publish books seem to disagree on a lot of critical details. In fact, the literature is so inconsistent that we are starting a public wiki-based documentation project.



In its basic form (non-HSPA), UMTS can deliver 384 kb/s per channel with up to 4 such channels active at once, assuming good link margins, small delay spreads, good power control, proper phase of moon, low traffic levels in surrounding cells, and generally clean living and happy thoughts on the part of everyone involved. That's just over 1.5 Mb/s on a really, really good day. To do this, UMTS musters just about all of the fancy math that the 1990's had to offer: orthogonal spreading codes (at least, orthogonal until they hit the real world), turbo codes, rake receivers, multisensor diversity demodulators. In exchange for all of this complexity you get roughly half (yes, half) the usable bit rate that you could get from an EDGE-style PSK/TDMA interface in the same bandwidth, except that the PSK/TDMA approach would provide more solid QoS guarantees, use a lot fewer transistors and be more power-efficient due to smaller crest factors in the amplifiers.

So, you might ask, why all of the new complexity if we get no performance advantage? Because simple approaches don't produce enough new intellectual property to keep the usual suspects in business. If you just take the GSM/GPRS/EDGE specification and change some parameters (like symbol rate and number of timeslots) you can get a very efficient, flexible system, but you won't have a much fodder for IEEE papers. Worse yet, you don't generate a lot of new patents, and with a lot of the patents on 2.xG systems expiring early in this decade, the NEPs needed a new gravy train. This complexity also deters small players from building their own UMTS implementations. (We have taken that as a challenge, but then we are unreasonable people.)

And what did UMTS do for the carriers? It nearly killed them. They overbid for the spectrum, barely had enough money left to roll out these really expensive new networks and when it was over they discovered that all their customers really wanted was better coverage and lower bills, neither of which have anything to do with UMTS. (The iPhone later proved that lack of demand for data was mostly due to the carriers' walled-garden approach to the itnernet, but that's fodder for a different post.) The killer app for the next five years turned out to be text messaging and economists got to write lots of papers saying things like "We conclude that the rationalization of bidding in the United Kingdom’s UMTS auction remains problematic." And just as the carriers are starting to recover from UMTS, the usual suspects start pushing the next shiny, new thing: LTE/IMS.


11 comments:

  1. Any thoughts on CDMA/EVDO thing?

    ReplyDelete
  2. I think CDMA is a bad choice for cellular air interfaces, regardless of the implementation. Cdma2000 WCDMA is just as bad as UMTS WCDMA for all of the same reasons. A single company has a stranglehold on the IP. The codes are never really orthogonal on real channels, so performance is never anywhere close to theoretical. It requires tight power control to prevent near/far dynamic range problems. It requires the radio to run at wide bandwidths and high duty cycles regardless of how much bandwidth the applications using that radio really need. Since the main performance limiter is inter-user interference, QoS in the lower layers varies wildly with system load. CDMA is just a bad technology for this application.

    ReplyDelete
  3. Hey David,
    Sent you a mail the other day about opportunities in Africa, can you please respond.
    thanks

    ReplyDelete
  4. Some things stated in this article might be true, hovever most networks which run HSPA today provide very good QoS... and they are just a software upgrade of the UMTS (Rel99)networks.

    So definitely UMTS had limitations but it was far from being a dead-end as a technology.

    With regards to the carriers spending way too much money in spectrum licenses, it was due to the internet bubble context, not to a specific technology...

    ReplyDelete
  5. Yes, granted, by doubling again the complexity of the PHY, UMTS can provide the same QoS of a much simpler system. And I say that UMTS is grotesquely complex for what it does, not that it is a dead-end technology. It is likely to be with us for another 15 years, maybe longer since a move to 4G requires a completely new core network.

    ReplyDelete
  6. David, HSPA+ provides very high bit rate and QoS, you cannot compare the first UMTS release with the latest GSM (EDGE) release... it's simply wrong!

    (The same is for HSPA+ and LTE, now they provide similar bandwith, but LTE will improve in a few years, WCDMA is at his max)

    Bye

    ReplyDelete
    Replies
    1. I'm not actually comparing UMTS to EDGE. I am saying that EDGE performance is an existence proof that it would have been possible to build a system with comparable performance without nearly so much complexity. I still stand by that. This stuff is all horribly over-engineered. You cannot possibly say that UMTS or HSPA+ is an elegant design. Furthermore, the spec appears to be deliberately obfuscated and many of the widely accepted books on these systems differ from both the spec and from the behavior of real systems.

      Really, are you saying that it would not be possible to achieve the same level of performance in a much simpler system? (Because from what I have heard, LTE *is* a simple system, although I have not yet started reading the specs for myself.)

      Delete
  7. I agree, CDMA is just the wrong technology. We did not really want this in Europe and voted for some OFDM system, but America/Qualcomm was able to convince us :-( Now we have a high speed system and Qualcomm chipsets like MDM9200 that still in their latest revisions are not able to handover a 42Mbps connection without big chances for a chipset reboot, we have big trouble covering a populated soccer stadium with reliable UMTS as interference is a killer in these crowded situations, we have mobile units that eat up the battery content every day, we have USB modems that can keep your coffee warm...

    But finally we got our OFDM, under the LTE label, and even in its earliest installations it outperformed the CDMA stuff, every day new stations pop up here in Germany. LTE fulfills what UMTS had promised!

    ReplyDelete
  8. Is Open BTS capable to support 3G? What will be its cost benefits over existing system?

    ReplyDelete
    Replies
    1. I think the big issue is the intellectual property over UMTS is probably still going to be a major hurtle....I mean IIRC you can only do the most basic GSM release before IP arguments might ensue and I'm not even talking about bare bones GPRS..I really want to see OpenBTS with UMTS, I don't even care about the latest releases or even HSDPA 1.8 but I'd like to see OpenBTS using UMTS in the 900MHz unlicensed band for example in the US by a WISP, could easily offer cellular phone service with the typical WISP high power WiFi. Only problem is UMTS needs 5MHz of spectrum. You could use 925-930 for downlink (900MHz ISM band in US is 902-928 though) and 910-915 for uplink. Not sure what is above 928MHz in the US but I imagine in rural area you could get the FCC to waive the restriction more easily then trying to temporarily lease spectrum from an incumbent carrier which requires two waivers (one from the carrier and one from the FCC) and of course an area where they aren't using the spectrum……

      Delete
  9. I posted this as part of a reply below but I figure itll get more attention as a general comment as well. I'd like to see OpenBTS using UMTS in the 900MHz unlicensed band for example in the US by a WISP, could easily offer cellular phone service with the typical WISP high power WiFi. Only problem is UMTS needs 5MHz of spectrum. You could use 925-930 for downlink (900MHz ISM band in US is 902-928 though) and 910-915 for uplink. Not sure what is above 928MHz in the US but I imagine in rural area you could get the FCC to waive the restriction more easily then trying to temporarily lease spectrum from an incumbent carrier which requires two waivers (one from the carrier and one from the FCC) and of course an area where they aren't using the spectrum……

    Anyone have any thoughts on using the 900MHz unlicensed band for UMTS on a test basis or even as a small WISP operator? I haven't seen any suggestions on this online and with the abundance of quad-band UMTS handsets you could easily get a bunch of people on the network without needing new handsets.

    ReplyDelete