From: XYZ on 10 Apr 2010 19:59 Hi, I've been struggling with OFDM timing synchronization using Schmidl algorithm for some time but I can't figure out how to find the sample at which the symbol starts. Authors seem to have the same problem but apparently they don't mention how to deal with it  or is it just me not understanding it? The timing metric shows me where to sample (I actually only use P metric because AGC term R gives false peaks for nontraining symbols) and I sometimes get my symbols circularly shifted. This happens when maximum of the metric is slightly off by one or two samples. For example after ifft of the first training symbol I transmit: first half second half cyclic prefix 1 2 3 4 5 1 2 3 4 5 1 2 3 > 0 1 2 3 4 5 6 7 8 9 10 11 12 sample number N but the optimal timing point is for N=1. Knowing that useful part of OFDM symbol is 10 samples I take 10 samples starting from the second one which gives 2 3 4 5 1 2 3 4 5 1 This sequence carries the same information but circularly shifted and performing fft on it does not give what was put into ifft at the transmitter. How can I find out that a shift occured and by how many samples? I can't simply try all possible shifts to find the correct offset because subcarriers are altered in the channel and I need to know this correct offset within the symbol to perform channel estimation by (circshift(symb_received,n) ./ symb_expected) I also tried to investigate Park algorithm that gives very sharp metric. Authors say that the proposed preamble is of the form Ppro = [C D C* D*] and it is possible to generate this pattern by transmittng a PN sequence of BPSK points on even frequencies and zeros on odd frequencies. So for example s = [1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0] should work but after ifft it simply gives Ppro2 = [C D C D]  no conjugates in the second half of the symbol. I wouldn't even expect it to yield something else. But then how to generate symbol of the form [C D C* D*]? Alternatively, please recommend another algorithms I could try. My channel is dispersive. Thanks!
From: Frank_os on 13 Apr 2010 12:09 >Hi, > >I've been struggling with OFDM timing synchronization using Schmidl >algorithm for some time but I can't figure out how to find the sample at >which the symbol starts. Authors seem to have the same problem but >apparently they don't mention how to deal with it  or is it just me >not understanding it? > >The timing metric shows me where to sample (I actually only use P metric >because AGC term R gives false peaks for nontraining symbols) and I >sometimes get my symbols circularly shifted. This happens when maximum >of the metric is slightly off by one or two samples. > >For example after ifft of the first training symbol I transmit: > > first half second half cyclic prefix > 1 2 3 4 5 1 2 3 4 5 1 2 3 >> > 0 1 2 3 4 5 6 7 8 9 10 11 12 sample number N > > >but the optimal timing point is for N=1. Knowing that useful part of >OFDM symbol is 10 samples I take 10 samples starting from the second one >which gives > > 2 3 4 5 1 2 3 4 5 1 > >This sequence carries the same information but circularly shifted and >performing fft on it does not give what was put into ifft at the >transmitter. > >How can I find out that a shift occured and by how many samples? I can't >simply try all possible shifts to find the correct offset because >subcarriers are altered in the channel and I need to know this correct >offset within the symbol to perform channel estimation by >(circshift(symb_received,n) ./ symb_expected) > >I also tried to investigate Park algorithm that gives very sharp metric. >Authors say that the proposed preamble is of the form Ppro = [C D C* D*] >and it is possible to generate this pattern by transmittng a PN sequence >of BPSK points on even frequencies and zeros on odd frequencies. So for >example > s = [1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0] should work but after ifft >it simply gives Ppro2 = [C D C D]  no conjugates in the second half of >the symbol. I wouldn't even expect it to yield something else. But then >how to generate symbol of the form [C D C* D*]? > >Alternatively, please recommend another algorithms I could try. My >channel is dispersive. > >Thanks! > Both Schmidl and Park algorithms assume the AWGN channel. The property of two identical symbols is destroyed by the dispersive channel. A better approach is to compute the cross correlation between the received signals and the preamble, but this approach increases the complexity of frame synchronization.

Pages: 1 Prev: 16 qam receiver symbol clock recovery noise Next: FMCW Radar help required. 