Differences

This shows you the differences between two versions of the page.

Link to this comparison view

projects:mkat_sband:pub:decfilters [2021/10/22 13:54]
wucknitz 		projects:mkat_sband:pub:decfilters [2022/06/15 10:49] (current)
wucknitz
Line 95: Line 95:
  
  
-For a comparison I ran the fits without the analog gains, then applied them after the fits, and finally ran fits with the analog gains included. This is with min-max for the relative aliasing and least-squares for the passpand:+For a comparison I ran the fits without the analog gains, and then the analog gains included. This is with min-max for the relative aliasing and least-squares for the passpand. The tradeoff plots do now include the rms ripple as well:
  
 +{{:projects:mkat_sband:pub:filtertradeoff_35_noanalog.pdf|tradeoff for fits without analog gain}} \\
 {{:projects:mkat_sband:pub:filtertradeoff_35_analog_beforefit.pdf|tradeoff for fits including analog gain}} {{:projects:mkat_sband:pub:filtertradeoff_35_analog_beforefit.pdf|tradeoff for fits including analog gain}}
 +
 +== Fit without analog gain ==
 +
 +Without the analog gain, there seems to be a sweet spot for 92% bandwidth near -50dB leakage. This is how the filter looks like (second plot shows passband ripple in more detail, on a log and linear scale, with total scale modification to bring all to the same mean level):
 +
 +{{:projects:mkat_sband:pub:filter_35_92per_noanalog_a.png?direct&550}} \\
 +{{:projects:mkat_sband:pub:filter_35_92per_noanalog_a_pass.png?direct&550}}
 +
 +But the analog gain does exist, and this is how the result looks like for the same filter, but with the analog gain included (not in the fit, only in the plots):
 +
 +{{:projects:mkat_sband:pub:filter_35_92per_analog_after_a.png?direct&550}} \\
 +{{:projects:mkat_sband:pub:filter_35_92per_analog_after_a_pass.png?direct&550}}
 +
 +Not so great, and it would be good to correct some of the analog gain with the decimation filter.
 +
 +== Fit with analog gain included ==
 +
 +For the fit including the analog gain, the min-max passband ripple cannot be reduced below 8.5dB, even for 90% bandwidth, which is a result of the low analog gain near 2.5 GHz.  
 +Looking at individual filters, one at 92% bandwidth with -50 dB aliasing is still looking acceptable. The rms passband ripple is about 1.2dB:
 +
 +{{:projects:mkat_sband:pub:filter_35_92per_analog_a.png?direct&550}} \\
 +{{:projects:mkat_sband:pub:filter_35_92per_analog_a_pass.png?direct&550}}
 +
 +On the linear scale, the large-scale passband ripple looks a bit scary, even though it is probably still better than the analog gain itself or the analog gain with the current filter. The large waves are exactly known and reproducible, but the general astronomer would not like to see them anyway. Can they be corrected in the F engine?
 +If not, we may have to reduce them.
 +
 +
 +
 +==== Taking into account the analog aliasing ====
 +
 +The analog filters produce their own aliasing near the band edges 1750 and 3500 MHz. Using the S21 coefficients of the IF processor we can compute the aliasing by comparing the true response with the one mirrored at 1750 and 3500 MHz:
 +
 +{{:projects:mkat_sband:pub:analog_aliasing.png?direct&600|analog aliasing}}
 +
 +The upper panel shows the response function in red and the mirrored versions in blue and green. The sum is shown in red dashed. For the lower panels the mirrored ones are divided by the true one (difference on dB scale) and the result is plotted only for the band edges. We reach -30 dB relative aliasing at 1916 and 3442 MHz and -45 dB at ca. 2019 and 3401 MHz. The vertical lines indicate a loss of 10/20/30 % of the bandwidth.
 +
 +We want to include the analog-aliasing in the total analog gain and exclude the range that is bad anyway from the optimisation of the digital filters. The limit for the latter is a bit arbitrary, let us try to exclude the > -30 dB range.
 +
 +
 +
 +
 +===== Final design (ver2) =====
 +
 +After consistency checks of a preliminary version (ver0) by Eddy, final filters were produced. Relative leakage was limited to -50dB (per frequency) including the effect of the analog bandpass. Near the edges the analog bandpass was cut with a lower limit to avoid upscaling too much. For the outer bands (LP and HP) the usable band was defined as 92%, for the bandpasses it is only 83%, because losses on both sides have to be accepted. This numbers were chosen to keep the passband ripple at a reasonable level, similar to the one already produced by the analog gain. For alternative filters we could trade passband flatness for usable bandwidth. With the available 35 coefficients, the following filters are optimal in this sense.
 +
 +Here is one example plot from the actual fitting process, showing where the analog gain curve was limited: \\
 +{{:projects:mkat_sband:pub:ver2_fitting.png?direct&550}}
 +
 +
 +=== S0 ===
 +
 +{{:projects:mkat_sband:pub:ver2_S0.pdf|All plots for S0 in one PDF file.}}
 +
 +== With analog gain (full band and passband) ==
 +
 +
 +{{:projects:mkat_sband:pub:ver2_S0_with_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S0_with_analog_passband.png?direct&400}}
 +
 +
 +== Without analog gain (full band and passband) ==
 +
 +{{:projects:mkat_sband:pub:ver2_S0_without_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S0_without_analog_passband.png?direct&400}}
 +
 +
 +
 +
 +
 +
 +=== S1 ===
 +
 +{{:projects:mkat_sband:pub:ver2_S1.pdf|All plots for S1 in one PDF file.}}
 +
 +== With analog gain (full band and passband) ==
 +
 +
 +{{:projects:mkat_sband:pub:ver2_S1_with_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S1_with_analog_passband.png?direct&400}}
 +
 +
 +== Without analog gain (full band and passband) ==
 +
 +{{:projects:mkat_sband:pub:ver2_S1_without_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S1_without_analog_passband.png?direct&400}}
 +
 +
 +
 +=== S2 ===
 +
 +{{:projects:mkat_sband:pub:ver2_S2.pdf|All plots for S2 in one PDF file.}}
 +
 +== With analog gain (full band and passband) ==
 +
 +
 +{{:projects:mkat_sband:pub:ver2_S2_with_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S2_with_analog_passband.png?direct&400}}
 +
 +
 +== Without analog gain (full band and passband) ==
 +
 +{{:projects:mkat_sband:pub:ver2_S2_without_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S2_without_analog_passband.png?direct&400}}
 +
 +
 +
 +=== S3 ===
 +
 +{{:projects:mkat_sband:pub:ver2_S3.pdf|All plots for S3 in one PDF file.}}
 +
 +== With analog gain (full band and passband) ==
 +
 +
 +{{:projects:mkat_sband:pub:ver2_S3_with_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S3_with_analog_passband.png?direct&400}}
 +
 +
 +== Without analog gain (full band and passband) ==
 +
 +{{:projects:mkat_sband:pub:ver2_S3_without_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S3_without_analog_passband.png?direct&400}}
 +
 +
 +
 +=== S4 ===
 +
 +{{:projects:mkat_sband:pub:ver2_S4.pdf|All plots for S4 in one PDF file.}}
 +
 +== With analog gain (full band and passband) ==
 +
 +
 +{{:projects:mkat_sband:pub:ver2_S4_with_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S4_with_analog_passband.png?direct&400}}
 +
 +
 +== Without analog gain (full band and passband) ==
 +
 +{{:projects:mkat_sband:pub:ver2_S4_without_analog_fullband.png?direct&400}}{{:projects:mkat_sband:pub:ver2_S4_without_analog_passband.png?direct&400}}
 +
 +
 + --- //Olaf Wucknitz 2022/01/26 11:36//
 +
 +
 +===== Old filters for comparison =====
 +
 +Here are the corresponding plots for the filters used until end of May 2022.
 +
 +{{:projects:mkat_sband:pub:orig_S0.pdf|All plots for S0 in one PDF file.}} \\
 +{{:projects:mkat_sband:pub:orig_S1.pdf|All plots for S1 in one PDF file.}} \\
 +{{:projects:mkat_sband:pub:orig_S2.pdf|All plots for S2 in one PDF file.}} \\
 +{{:projects:mkat_sband:pub:orig_S3.pdf|All plots for S3 in one PDF file.}} \\
 +{{:projects:mkat_sband:pub:orig_S4.pdf|All plots for S4 in one PDF file.}}
 +
 +
 +
 +===== Test observations =====
 +
 +The new coefficients were tested in five 10-min scans, one per subband, of J1939-6342 on 8th June 2022. Here are combined results for all subbands. For each subband we have one observation with old and one with new filter. The new ones have timestamps in the title that start with 1654..., old ones are smaller. Note that not all antennas participated in all observations.
 +
 +{{:projects:mkat_sband:pub:sefd_filtertest.pdf|PDF with SEFD curves}} \\
 +{{:projects:mkat_sband:pub:bandpass_filtertest.pdf|PDF with bandpass curves}}\\
 +{{:projects:mkat_sband:pub:bandpass_corr_filtertest.pdf|PDF with bandpass curves, amplitude corrected for digital decimation filter for a better comparison}}\\
 +{{:projects:mkat_sband:pub:sefd_bandpass_filtertest.pdf|PDF with SEFD and bandpass curves (dB)}}\\
 +{{:projects:mkat_sband:pub:sefd_bandpass_corr_filtertest.pdf|PDF with SEFD and bandpass curves (dB), amplitude corrected for digital decimation filter for a better comparison}}
  
 
projects/mkat_sband/pub/decfilters.1634903694.txt.gz · Last modified: 2021/10/22 13:54 by wucknitz     Back to top

Impressum / Imprint | Datenschutzhinweis / Privacy Policy