WARP Project Forums - Wireless Open-Access Research Platform

You are not logged in.

#1 2020-Sep-10 01:29:32

alex
Member
Registered: 2019-Apr-27
Posts: 2

weird channel estimation?

Hi,

I connect two WARPv3 by coaxial cables to see how cable length affects the channel response. I'm expecting that a longer cable will give a steeper phase slope in the freq. domain. However, as I increase the cable length, the slope gets flatter, which means the signal experiences a shorter delay. More specifically, as the cable length increases by ~47 inch, the delay *decreases* by 3ns. This doesn't make sense.

I'm running wl_example_siso_ofdm_txrx.m, sending from RFA on one board to RFA on the other, and two WARPs are synchronized by CM-PLL clock modules. I can't figure out what might have possibly gone wrong here? Any comments or suggestions are appreciated. Please help. Thank you.

Alex

Offline

 

#2 2020-Sep-10 22:09:12

murphpo
Administrator
From: Mango Communications
Registered: 2006-Jul-03
Posts: 5140

Re: weird channel estimation?

How are you estimating delay from the channel estimate phase (i.e. what processing do you apply to rx_H_est)?

Depending on your estimation scheme, you may be observing the delay estimate wrapping due to crossing a sample period boundary. Assuming propagation speed of 0.5c in coax (very rough estimate), 4 ft takes ~8nsec, a large fraction of the 25nsec sample period.

Offline

 

#3 2020-Sep-10 22:21:32

alex
Member
Registered: 2019-Apr-27
Posts: 2

Re: weird channel estimation?

I simply correlate rx_H_est with delay responses, exp(-1j*2*pi*delay/64*[0:63]), and find the peak. With a 47 inch and 94 inch cable, I find the peak at 10ns and 7ns, respectively. Also, FFT_OFFSET variable is set to 0 and AGC is disabled. Since the WARPs are synchronized by CM-PLL, I get exactly the same results every time.

You mentioned estimate wrapping. If I do ifft on rx_H_est, I do see energy spread over to 60-64 taps due to wrapping but the peak is still within the first tap.

Last edited by alex (2020-Sep-10 22:49:07)

Offline

 

#4 2020-Sep-11 11:21:53

murphpo
Administrator
From: Mango Communications
Registered: 2006-Jul-03
Posts: 5140

Re: weird channel estimation?

correlate rx_H_est with delay responses, exp(-1j*2*pi*delay/64*[0:63])

At a high level this looks right, but how you define the "delay" parameter is important. I find it helpful to think of delay as a frequency-dependent phase. Specifically, delay manifests as a sloped phase across frequency where the slope is proportional to the delay. However in the OFDM Tx/Rx example the rx_H_est is derived from the preamble after the preamble itself was used to synchronize the Rx processing. The LTS correlation process chooses the Rx sample period with the highest correlation peak, defines this as the "zero" sample for the FFT_OFFSET adjustment, then uses that sample alignment to take the FFT of each OFDM symbol. So rx_H_est can only be used to estimate sub-sample delay - any delay component larger than a sample period is unknowable by the LTS correlation / channel estimation process. Thus, if you induce a delay that moves across sample period boundaries, the delay estimate based on rx_H_est will wrap, only showing the new sub-sample (fractional) delay component.

Offline

 

Board footer