Zadoff-Chu correlation with frequency offset
in [DSP]
|
Prev: Call for Papers: The 2010 International Conference on Genetic and Evolutionary Methods (GEM'10), USA, July 2010
Next: Any faster method to apply convolution?
From: Nitram on 8 Dec 2009 21:24 Hi, I have an issue with Zadoff-Chu codes I am trying to understand. I generate a ZC sequence, remove the center carrier (as per the LTE spec) and try to correlate it in the time domain relative to a frequency shifted version of itself. Finally, I plot the correlation. Here's the corresponding Matlab code. u = 25; for n=1:63 zc(n) = exp(-1i*pi*u*(n-1)*n/63); end zc(32) = 0; freqShift = 1; zcShifted = circshift(zc, freqShift); timeCorr = xcorr(fft(zc, 2048), fft(zc, 2048)); plot(abs(timeCorr)); My issue is that I still detect a correlation peak, albeit at different time offset (Even worse, depending on the frequency offset, I can detect a peak at exactly the same time as I would for a 0 frequency shift). Is this behavior expected, or am I doing something wrong? In the unlikely case this is expected, in a scenario where both frequency shifts and timing offset can happen, how can I discriminate between the two when detecting my ZC sequence? Thank you so much.
From: Vladimir Vassilevsky on 8 Dec 2009 23:09 Nitram wrote: > Hi, > I have an issue with Zadoff-Chu codes I am trying to understand. I > generate a ZC sequence, remove the center carrier (as per the LTE spec) and > try to correlate it in the time domain relative to a frequency shifted > version of itself. Finally, I plot the correlation. > My issue is that I still detect a correlation peak, albeit at different > time offset (Even worse, depending on the frequency offset, I can detect a > peak at exactly the same time as I would for a 0 frequency shift). Is this > behavior expected, or am I doing something wrong? This is what expected. Time/frequency ambiguity function is classic problem. > In the unlikely case this is expected, in a scenario where both frequency > shifts and timing offset can happen, how can I discriminate between the two > when detecting my ZC sequence? Either acquire frequency first or use incoherent detectors to lock the timing. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
From: Nitram on 9 Dec 2009 08:40
Thank you for the prompt reply. The trouble is that I was hoping to use this method for coherent detection: first scan in time for a peak and if no peak is found, shift the frequency and try the time scan again. Here, not only do I get a peak with the wrong frequency offset but also at the wrong time. Do you have different time-domain incoherent detectors you can recommend (I already played with a frequency-domain incoherent detector that does seem to work). Thanks > >Nitram wrote: >> Hi, >> I have an issue with Zadoff-Chu codes I am trying to understand. I >> generate a ZC sequence, remove the center carrier (as per the LTE spec) and >> try to correlate it in the time domain relative to a frequency shifted >> version of itself. Finally, I plot the correlation. >> My issue is that I still detect a correlation peak, albeit at different >> time offset (Even worse, depending on the frequency offset, I can detect a >> peak at exactly the same time as I would for a 0 frequency shift). Is this >> behavior expected, or am I doing something wrong? > >This is what expected. Time/frequency ambiguity function is classic problem. > >> In the unlikely case this is expected, in a scenario where both frequency >> shifts and timing offset can happen, how can I discriminate between the two >> when detecting my ZC sequence? > >Either acquire frequency first or use incoherent detectors to lock the >timing. > > >Vladimir Vassilevsky >DSP and Mixed Signal Design Consultant >http://www.abvolt.com > |