I have a white noise data for 10 sec ( with 2 min silence or zero data at start and end)that is captured by sensor after delay t1( which I dont know). same data is captured after delay t1 and t2 by sensor 2. I want to compute transfer function(system identification) between sensor 1 and sensor 2. when I use LMS( I know unknown is IIR) to adapt, I get filter which has zeros outside the unit circle, and impulse response which is non symmetric. I am told that this is typical of non-causal filter.
- If I use sensor 1 data, capture portion of say 5 sec of white noise data from the first non zero sample, treat it as my input , and then do the same for sensor 2 data , treating it as desired signal for LMS, would this be a right approach?
- Alternatively, assuming a group delay of filter_len/2, where filter_len is the length of LMS filter, sensor 2 data would be delayed by this amount , with respect to sensor 1. So, capturing white noise data starting from first non zero sample would amount to adapting the filter too soon( disregarding the group delay) and hence will give anticausal filter?
I tried capturing sensor 2 starting from group delay samples before the non zero sample, and repeat the LMS adaption, but the filler still looks all over the place.( see Figure attached , with LMS of filter length 256, step size 0.08) Any suggestions on how can I solve this problem?