From dec7fd0994b97443f79b84790e08c89a370cbd24 Mon Sep 17 00:00:00 2001 From: Eric Kooistra <kooistra@astron.nl> Date: Thu, 15 Dec 2022 08:54:02 +0100 Subject: [PATCH] Keep anti aliasing filter notes from APERTIF. --- .../dsp/doc/anti_aliasing_filter_notulen.txt | 64 +++++++++++++++++++ 1 file changed, 64 insertions(+) create mode 100755 libraries/dsp/doc/anti_aliasing_filter_notulen.txt diff --git a/libraries/dsp/doc/anti_aliasing_filter_notulen.txt b/libraries/dsp/doc/anti_aliasing_filter_notulen.txt new file mode 100755 index 0000000000..6087e7ca74 --- /dev/null +++ b/libraries/dsp/doc/anti_aliasing_filter_notulen.txt @@ -0,0 +1,64 @@ +Apertif Redmine #709 anti-aliasing filter + +Problem: +The subband filterbank (Fsub) is critically sampled. This causes aliasing at the edges of the subbands +and that shows in the channels that are created by subsequent the channel filterbank (Fchan). The +Matlab two_pfb.m simulation of the Fsub and Fchan shows this by simulating a chirp input through the +frequency band: + + https://svn.astron.nl/UniBoard_FP7/RadioHDL/trunk/applications/apertif/matlab/two_pfb.m + +Erik van der Wal has invented a solution that can compensate for this aliasing. This we refer to as +Anti Aliasing Filter (AAF). Stefan has cleaned up Erik's original Matlab code: + + https://svn.astron.nl/UniBoard_FP7/RadioHDL/trunk/applications/apertif/matlab/unripple/unripple_v2_matlab.m + +into: + + https://svn.astron.nl/UniBoard_FP7/RadioHDL/trunk/applications/apertif/matlab/unripple/AAF_APERTIF.m + +Bas can spend time on this, Stefan is too busy with other projects. Action Eric: discuss with Bas and +Stefan. Agnes has set up a meeting between Eric, Bas and Stefan in week 24 - 2017. + +Discussion with Bas and Stefan is to clarify: + +1) what will be gained scientifically by avoiding or solving the problem? + - how does this relate to the L0 requirements of Apertif and Arts + +2) identify possible solutions: + a) flag and ignore the affected channels (Lofar approach) + b) avoid the problem by using an oversampled subband filterbank (SKA Low CBF approach) + - the oversampling factor increases the processing load for Fsub, BF and Fchan + - the oversampling factor also increases the load on the fiber links, which are already fully loaded + c) correct the affected channels using the AAF + +3) does this AAF algorithm work in the case of APERTIF? + - with beamforming and fringe stopping in between Fsub and Fchan + +4) where in the processing chain should this be implemented + - for APERTIF correlator in firmware after the channel filterbank or in software after integration on the visibilities + - for ARTS tied array beamforming + +5) can it be implemented + - hardware limitations like processing power, are the neighbouring channels available at the node + +Action points: +- SJW : Start "ASTRON_RP_1549_Apertif_Anti_Aliasing_Filter.tex" (AAAF) document in Latex to document our ongoing work +- SJW/BT : Obtain relevant spectra from astronomers, apply the aliasing and then check the impact +- BT : Get familiar with AAF alorithm and add AAF description to the AAAF document +- EK : Organize that Stefan and Bas get access to Apertif firmware repository in SVN and to Apertif Redmine for AAF issue +- EK : Add overview of possible solutions and Apertif/Arts processing chain to the AAAF document +- EK : Update AAF redmine issue #709 + + + Apertif BF @ dish Apertif X in central building + + ADC --> Fsub --> BF ---> fiber links ---+---> Fchan --+--> Correlator (X) --> network --> Data writer + compound beam data | | visibilities + | | + | | Arts BF for SC3,4 in central building + | +--> Tied array BF --> network --> Transient Search pipeline (SC3,4) + | IUQV TABs + | + | Arts BF for SC1 and SC2 in the central building + +--> Tied array BF --> network --> Pulsar Timing (SC1), VLBI (SC2) \ No newline at end of file -- GitLab