Slightly technical, but the most qualified and credible analysis of the reasons that S-CDMA will not become a standard:
G Gilder (or his writers) believe highly in CDMA technology, as demonstrated by their numerous comparisons to Qualcomm, and the state of Terayon vs standards vs etc. When Gilder published that article several months ago, there are a number of hype items and inaccuracies. If the article were written without these, it would be more of an even play. For instance, the article fails to highlight the reduction in data carrying capacity when S-CDMA adjusts for increasing noise. This reduction greatly effects the number of active cable modems and their subsequent quality of service. Another example is that TERN's current S-CDMA provides symmetric downstream and upstream data rates, while the world (DOCSIS, DAVIC, Com21, Moto) is running at 30 Mbps (raw) downstream with multiple upstream channels. With the inherently asymmetric traffic load imposed by TCP traffic and symmetric channels, you'll run out of downstream before you run out of upstream; i.e., you need a higher capacity downstream (as compared to upstream) at this time.
Another way to think of this, is for every TERN or LANcity (now Nortel) downstream channel that is installed, the cable operator is throwing away between approx 15 Mbps (raw, TERN) to 20 Mbps (raw, LANcity) of usuable data capacity. Note the current TERN S-CDMA system is not forward compatible (upgradable) to DOCSIS, as there is no 64 QAM support.
S-CDMA has it's uses. It does "mine" data capacity out of RF space on the upstream that might be unusable by QPSK. However, it's likely that for S-CDMA to work running most of its spreading codes, it means the spectrum is cleaner. If it's cleaner, then it's cheaper and more effective (in my opinion) to run QPSK with good forward error correction. I've seen ideal (analytical) performance curves of QPSK and 16QAM with varying amounts of forward error correction. What I've not seen published, is S-CDMA performance on these same curves. Discussions I had with folks during IEEE 802.14a suggest that the complexity of S-CDMA only buys 2-3 dB of noise floor performance gain. That is, that S-CDMA will perform with equal bit error performance in 2-3 dB "dirtier" upstreams. On some cable plants, that 2-3 dB may be the difference between running any high-speed data service or not.
However, cable plant noise floor can vary on a daily basis or with variation in other factors, such as unknown noise sources, changes in equipment adjustment, etc. Cable operators, who are aiming at two-way, target 25 dB CNR as the minimum headroom. They try to get up as high as 40 dB. 20+dB is sufficient for QPSK with good forward error correction and some implementations will run at full packet rate capacity down to 12.9 dB (e.g. Com21's). S-CDMA can go 2-3 dB "lower" with same error performance. It can also go further, but when it does, it trades capacity for error performance. Please note that cable plants are highly "creative" in the types of noise that they can accept and mileage will vary. There may be some types of narrow band interference that S-CDMA copes with well while QPSK packet error rate may be effected. The opposite may be true also.
The real point is that cable operator's have to manage the noise floor in the upstream to acceptable limits for *all* services they may be running (e.g., high speed data cable modems, TV channels, remote management, voice, interactive set top boxes, etc.) It's likely that QPSK (with good forward error correction) will work just fine. However, if you buy into the FUD, then you'll start looking at S-CDMA with more interest as it may be akin to buying insurance if there's any uncertainty about the plant.
---------------
Exerpts from Message 11392112
Author was Chief Technology Officer of Com21 when written, but technical shortcomings should not be dismissed by corporate affiliation. |
