2
$\begingroup$

I want to estimate a channel based on LTE 3GPP EVA with given power delay profile (set of average power and delay of channel taps).

tau = [0, 30e-9, 150e-9, 310e-9,  370e-9,  710e-9, 1090e-9,  1730e-9, 2510e-9]; % relative delay (s)
pdb = [0,  -1.5,   -1.4,   -3.6,    -0.6,    -9.1,    -7.0,    -12.0,   -16.9]; % avg. power (dB)

Unfortunately, the relative delays of channel taps are not multiples of my sampling time Ts=3.3333e-07 s. Here, I assumed the number of taps to be estimated is Ds/Ts, where Ds is the delay spread.

First of all, please let me know if there is anything wrong in my approach. Also, How should I measure my estimation error in this case? Should I compare my estimates with channel sampled at my sampling rate, or should I subtract the reconstructed frequency response of estimate and the original one? Anyhow, due to the difference in the sampling time and delays, there is an intrinsic error in my estimation no matter how I measure it.

Thanks in advance :)

$\endgroup$
  • 1
    $\begingroup$ The power delay profile is much more finely spaced than your sample interval. You just won't be able to express that level of detail in the channel model with your sample rate. You could simulate at a higher sample rate instead, or just resample the power delay profile to your lower sample rate as an approximation. As far as how to measure your estimation error, I think the most meaningful metric is, given a particular input signal, the difference between the channel's output and your channel estimate's output, probably in a squared-error fashion. $\endgroup$ – Jason R Mar 19 '13 at 0:51
  • $\begingroup$ Thanks Jason. Currently I'm implementing channel using MATLAB function h = rayleighchan(ts,fd,tau,pdb); where ts is the sampling time , fd is the doppler freq. ( for simplicity let assume fd=0 and static channel) and tau and pdb are the delay and average powers. $\endgroup$ – Siavash Mar 19 '13 at 17:24
  • $\begingroup$ Also, in real life, there are some channel paths that their delay are not multiples of sampling time. $\endgroup$ – Siavash Mar 19 '13 at 17:28
  • $\begingroup$ Regarding the estimation error, I totally agree with the metric that you proposed (RMSE or MSE). But, I don't have a point of reference as I don't have the channel Z transform with my sampling time. $\endgroup$ – Siavash Mar 19 '13 at 17:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.