From flind at haystack.mit.edu Fri Mar 25 14:26:19 2011 From: flind at haystack.mit.edu (Frank D. Lind) Date: Fri, 25 Mar 2011 14:26:19 -0400 Subject: [OpenRadar-developers] Echotek Filters Message-ID: <4D8CDE4B.2030004@haystack.mit.edu> Hi, There has been a lot of discussion about needing better RF filters for the Echotek digital receivers. In particular this is specific to the Graychip 4016 which is used on these boards. We have now put together and tested some filters using the following Mathworks example as a basis : http://www.mathworks.com/products/filterdesign/demos.html?file=/products/demos/shipping/filterdesign/ddcfilterchaindemo.html This of course requires matlab and the filter design toolkit to develop your own filters. Mathworks also provides nice methods visualize the filters and to pull useful information out of the cascaded filter chains (e.g. impz for the impulse response). It was necessary to bypass the resampler in the Graychip due to it being difficult to model properly. This corresponds to a specific set of register configurations to accomplish this on the hardware. The default filter distributed with the MIT Echotek driver has fairly poor performance and the newer filters correct this problem. You can find documentation of the scripts used for these filters, the resulting INI files, and some plots on the Open Radar site : http://www.openradar.org/cgi-bin/openradar.cgi/Projects/EchotekHardware/EchotekHardware/GC4016Filter Correspondence between the simulations and RF data taken with a noise diode source is good within the limits of the measurements. When using this script the filters are specified as half band (i.e. RF bandwidth is 2x specified bandwidth). Additionally if the stop band is beyond the nyquist rate the results on the actual hardware may not match the simulations. This is the likely cause of the passband shape seen with the default filter. To achieve steep rejection it is necessary to have the final RF bandwidth less than the final sample bandwidth. I recommend changing from the default filter to one which provides better performance. Regards, -- Frank -------------- next part -------------- A non-text attachment was scrubbed... Name: flind.vcf Type: text/x-vcard Size: 333 bytes Desc: not available URL: