From 0b7b6c09527e28c7a077b019170b98c7fc487b37 Mon Sep 17 00:00:00 2001
From: Eric Kooistra <kooistra@astron.nl>
Date: Wed, 11 Jan 2023 16:24:41 +0100
Subject: [PATCH] Clarify func_wpfb_subband_scale_w().
---
libraries/dsp/wpfb/src/vhdl/wpfb_pkg.vhd | 22 +++++++++++++++-------
1 file changed, 15 insertions(+), 7 deletions(-)
diff --git a/libraries/dsp/wpfb/src/vhdl/wpfb_pkg.vhd b/libraries/dsp/wpfb/src/vhdl/wpfb_pkg.vhd
index b175a1a1cf..54aea050a7 100644
--- a/libraries/dsp/wpfb/src/vhdl/wpfb_pkg.vhd
+++ b/libraries/dsp/wpfb/src/vhdl/wpfb_pkg.vhd
@@ -82,7 +82,8 @@ package wpfb_pkg is
-- LOFAR2 subband filter
-----------------------------------------------------------------------------
- -- Can use guard_w = 1, because fil_backoff_w = 1 already provides sufficient input margin
+ -- Use guard_w = 1, instead of 2 to avoid overflow in first FFT stage,
+ -- because fil_backoff_w = 1 already provides sufficient FFT input margin.
-- Fsub settings:
-- . Settings used on LTS and DTS until at least March 2022
@@ -99,7 +100,7 @@ package wpfb_pkg is
true, false, true, 23, 18, 1, 24, 1, true, 54, 2, 195313,
c_fft_pipeline, c_fft_pipeline, c_fil_ppf_pipeline);
- -- . Settings used in tb_tb_verify_pfb_wg with fft_out_dat_w = 19b, to preserve FFT processing gain of 4.5 bits
+ -- . Settings used in tb_tb_verify_pfb_wg with fft_out_dat_w = 19b, to preserve FFT processing gain of 5 bits
-- - use stage_dat_w = 25 --> fil_out_dat_w = fft_in_dat_w = 24
-- - with fft_out_dat_w = 19 --> stat_data_w = 2*19 + 18 = 56 b
constant c_wpfb_lofar2_subbands_dts_19b : t_wpfb := (1, 1024, 0, 6,
@@ -117,10 +118,16 @@ package wpfb_pkg is
-- The FFT output has more bits to be able to preserve the sensitivity of
-- the processing gain of the FFT. The FFT has a processing gain of
- -- sqrt(N_sub = N_fft / 2 = 512), so 4.5 bits. Therefore choose
- -- fft_out_dat_w = fil_in_dat_w + 5 = 14 + 5 = 19b. Using fft_out_gain_w =
- -- 1 compensates for the fil_backoff_w = 1 of the FIR filter. The
- -- func_wpfb_subband_scale_w then thus returns 19 + 1 - (14 + 1) = 5 bits.
+ -- W_fft_proc = sqrt(N_fft = 1024), so 5 bits. Therefore choose
+ -- fft_out_dat_w = fil_in_dat_w + 5 = 14 + 5 = 19b. Using fft_out_gain_w
+ -- = 1 compensates for the fil_backoff_w = 1 of the FIR filter.
+ -- However, instead keep fft_out_dat_w = 18b to fit a 18x19 multiplier in
+ -- the SST. To preserve the sensitivity increase fft_out_gain_w by 1 at the
+ -- expense of loosing factor 2 (1 bit) in subband dynamic range. Therefore
+ -- fft_out_gain_w = 2 and the func_wpfb_subband_scale_w() then thus returns
+ -- (fft_out_dat_w + fft_out_gain_w) - (fil_in_dat_w + fil_backoff_w) =
+ -- (18 + 2) - (14 + 1) = 5 bits = W_fft_proc, to preserve the subband
+ -- sensitivity.
function func_wpfb_subband_scale_w(wpfb : t_wpfb) return natural;
-- The WPFB subband gain is the expected factor between subband amplitude
@@ -130,11 +137,12 @@ package wpfb_pkg is
-- . DC gain of the FIR filter (= fir_filter_dc_gain ~= 1.0),
-- . the FFT gain for a real input (= c_fft_real_input_gain_sine = 0.5) and
-- . the extra bits to preserve the sensitivity of the FFT processing gain
- -- (derived from wpfb).
+ -- W_fft_proc = 5b (derived from wpfb by func_wpfb_subband_scale_w()).
-- For example:
-- . func_wpfb_subband_gain() ~= 8 for c_wpfb_lofar2_subbands_lts_2021 and
-- for c_wpfb_lofar2_subbands_dts_18b
-- . func_wpfb_subband_gain() ~= 16 for c_wpfb_lofar2_subbands_dts_19b
+ -- for c_wpfb_lofar2_subbands_l2ts_18b
function func_wpfb_subband_gain(wpfb : t_wpfb; fir_filter_dc_gain : real) return real;
-- The expected WPFB SST level for subband amplitude A_sub and an integration
--
GitLab