|
From: | Marcus Müller |
Subject: | Re: [Discuss-gnuradio] Recover the signal |
Date: | Tue, 27 Oct 2015 17:53:38 +0100 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.7.0 |
Ah, so it's FFT of samples 0-127, then FFT of samples 128-255, then
FFT of samples 256-383 and so on! That has an interesting consequence: first of all, the (I)DFT of a Zadoff-Chu sequence is, in fact, also a ZC sequence. However, that doesn't mean that the DFT of a Zadoff-Chu sequence multiplied with a rectangular Window, is one, too: let be your Zadoff-Chu sequence, and let be your FFT length; judging from your figures, you don't actually use all carriers; let's say there are used carriers. The following Fourier transform equalities hold: , which also is a Zadoff-Chu sequence (i.e. magnitude ), With your OFDM symbol being but the central out of carriers, I'd model the input of the IDFT as , which doesn't have Zadoff-Chu properties, so your signal in time domain will not have constant amplitude. In theory, the inverse operation in the receiver should give you the same , maybe having a bit of noise, and maybe having a constant DC offset if any, IFF transmitter and receiver were synchronized. What I assume is that the DC offset that you're seeing in the center peak is always the same, but the phase of the symbol start isn't; that leads to the DC offset always having the same phase, but the center symbol always a different one; if DC offset and center symbol happen to have the same phase, the peak is high, if they happen to oppose each other, you get cancellation. Generally, depending on your bandwidth you might do something called offset tuning: when you use "uhd.tune_request(,)" instead of directly using in the tuning frequency of a USRP block, you can use the USRPs capability to digitally tune by in the DSP, moving the DC offset / LO leakage out of band. Best regards, Marcus On 10/27/2015 12:15 PM, scott tiger
wrote:
|
[Prev in Thread] | Current Thread | [Next in Thread] |