The BER curves donot follow exactly as per the ones published in the book i am referring to, however they do resemble them. moreover the performance after using alamouti scheme does increases but not significantly as per the book. i have attached the pics of simulated curves. they were taken with 3.5K no of symbols, 10Hz symbol rate, 1 Hz doppler spread, 8 samples per symbol and mean of 20 iterations per SNR value. These values gives a very good result but the time taken for it outrageously large.

I have tried all the basic things i could, for example- using reenterant VIs, closing all the subVIs & block diagrams, minimal use of arrays. I feel that the problem is not of memory but more of computational time. Large computational time can cause very serious issues ahead in the project; i.e. when actual likelihood function will be implemented which will involve checking for all upto 32 symbols (upto 256 in QAM) in PSK for each 3.5K symbols. Decreasing the no of symbols will undoubtedly decrease the computational time but the desired BER resolution wont be possible (ie minimum is around 1E-5 for 3.5K symbols).

Theoretical Result:


Simulated Results:

  1. without Alamouti Scheme, without channel knowledge 
  2. without Alamouti Scheme, with channel knowledge  
  3. using Alamouti Scheme