Is HSDPA throughput all fairy tale ?

HSDPA is a 3GPP release 5 feature which promises a maximum of 14.4 Mbps per cell.
There is great confusion as to what this throughput really means and to what extent it can be utilised.
I've seen vendors and operators alike push for overblown node-b designs that can connect to multiple E1 links in order to carry all the HSDPA traffic and then sit back and watch the green bucks roll in the bank account.

In release 99 there are 16 phycial channels of SF 16. EAch of these channels carry a raw bit rate (uncoded) of 480kbps. Assuming that 15 out of 16 channels are used for traffic, a total of 7.2Mbps of raw data can theoretically be carried in one cell. Assuming 1/2 rate coding, this drops to 3.6Mbps. This calculation uses the number of orhogonal channels as a limit to throughput. In reality, the existence of interference and variability of mobile locations causes this maximum rate to drop tremendously. A typical release 99 UMTS cell can carry a total throughput of no more than 1 Mbps in the best conditions (low interference).

Let us compare the above with HSDPA.
HSDPA channels use fixed SF16 channels. They use adaptive modulation and coding, in the best case they can operate at 16 QAM without any coding, which brings the rate per SF16 channel to 960kbps and the total per 15 codes to 14.4 Mbps (one code used for other control and common channels).

It is ofcourse very unrealistic to assume that all mobiles will be in a favourable location and can use all 15 codes without having any coding and at the maximum modulation scheme.

All things equal, one should expect the average coding in the cell to be a number <1, may be clsoe to 1/2. The interference limited the throughput per cell in the R99 case to 28% (in the best case) of what is possible, there is no reason to think this will be different in the case of HSDPA. This brings the average throughput per cell down to around 2Mbps.

The scheduling algorithms used to schedule traffic on the HSDPA shared channel also impact the performance. If mobiles in good locations are always given priority, the total throughput per cell will increase above 2Mbps, but this is on the expense of faireness. If traffic is allocated to mobiles depending on how long they waited to get the traffic, then the throughput will be slightly less than 2Mbps.

I conclude that HSDPA will roughly double the average cell throughput compared to release 99. So if your R99 cell carries 500kbps, don't expect to get more than 1Mbps with HSDPA on average.