Hi, I have a problem in channel estimation of a waveform with the following specs: Baud rate = 2400 Sample rate = 9600 Hz USRP input sample rate = 400 kHz (three times interpolating by 5, one time decimating by 3) Super frame length = about 9 seconds (CFO removal at start of each) Channel estimation = around every 100 msec (training seq. space) Daughter board = LFTX / LFRX The problem is this: In some superframes, from each training seq. to the next, there is a jump in derivatives of the estimated phases (equivalent to jumps in carrier frequency) at the training seq. causes fatal errors in channel estimation. Increasing the carrier frequency makes it worse. I really wonder why that happens and how to treat it. () (I have tested the waveform with some radios and it worked well and I have been working on it for about 2 months...) Regards, Al