|
From: | Kyle Zhou |
Subject: | Re: [Discuss-gnuradio] wbx usrp1 sampling spectrum distortion |
Date: | Wed, 21 Apr 2010 10:28:00 +1000 |
User-agent: | Thunderbird 2.0.0.24 (Windows/20100228) |
Matt Ettus wrote:
On 04/20/2010 02:51 AM, Kyle Zhou wrote:Hi I use WBX + USRP1 to capture a DAB+ digital radio broadcast signal at frequency 204.64MHz and bandwidth 1.4MHz. Sampling rate is 2Msps (decimation=32) The spectrum shown by usrp_fft.py is normal and clean. However, after capturing the data using usrp_rx_cfile.py and loading the data to Matlab and show the spectrum, I found an unexpected portion of spectrum around the Nyquist frequency (1MHz in this case). The parameters I use for usrp_rx_cfile and usrp_fft are the same: -f 204.64M -d 32 -g 6. I uploaded a snapshot of the spectrum:http://lh4.ggpht.com/_ozkfEhDY6Mg/S81p3QihHkI/AAAAAAAAAsE/KJ8vtRo4EI4/spec_samp.jpgI can filter out the undesired part easily but that cost computational resource and I do want to find out the cause. Regards Kyle2 MS/s complex should give a spectrum of -1 to +1 MHz. Your Matlab plot show 0 to 2 MHz, so I think you are loading the samples as reals instead of floats.Matt
Thanks Matt.But I can confirm that I loaded complex data into Matlab using read_cshort_binary() provided by Gnuradio. Then I used pwelch() function to produce the power spectrum density estimate. According to the Matlab documentation, this function does accept complex data. It is because the frequencies at which the PSD is estimated is always in [0,Fsampling), the spectrum shows 0 to 2MHz, and I think 2MHz is equivalent to -1MHz, isn't it? I check the usrp_fft.py spectrum today again, using a sampling rate of 4M, now I can see there are adjacent channels at both sides. So I think the undesired spectral portion is from these neighbor channels.
Regards Kyle
[Prev in Thread] | Current Thread | [Next in Thread] |