diff --git a/applications/apertif/designs/apertif_unb1_correlator/tb/vhdl/tb_node_apertif_unb1_correlator_processing_output.vhd b/applications/apertif/designs/apertif_unb1_correlator/tb/vhdl/tb_node_apertif_unb1_correlator_processing_output.vhd
index 056a77b791c8027aa0852bf3f55349900f6b347d..9349f8312a3889ad63e6e15b3fb7c2eb43a7bd87 100644
--- a/applications/apertif/designs/apertif_unb1_correlator/tb/vhdl/tb_node_apertif_unb1_correlator_processing_output.vhd
+++ b/applications/apertif/designs/apertif_unb1_correlator/tb/vhdl/tb_node_apertif_unb1_correlator_processing_output.vhd
@@ -25,6 +25,26 @@
-- and DB in node_apertif_unb1_correlator_output
--
-- Desription:
+-- A) This tb verifies:
+--
+-- 1) p_check_inter_channel_delay: Verify that g_inter_channel_delay is in min
+-- max range
+-- 2) p_mm_stimuli : set offload src MAC and dest MAC
+-- p_mm_diag_data_buffer_output : verify offload src MAC, dest MAC, and
+-- BSN timestamp, channel and beamlet
+-- 3) p_mm_stimuli : set phasor waveform into BG and start BG
+-- 3a)p_verify_cor_src_out: verify expected channel data at cor_src_out
+-- interface for g_use_prefilter, g_use_prefilter_ones
+-- 3b)p_mm_diag_data_buffer_output :verify expected channel data at offload
+-- interface for g_use_prefilter, g_use_prefilter_ones
+-- 4) u_eth_statistics : verify number of offloaded visibility packet data
+--
+-- * Test 3a) is also covered by tb_node_apertif_unb1_correlator.vhd.
+-- * Test 2) is also covered by tb_apertif_unb1_correlator_nodes.vhd using
+-- file IO.
+-- * Test 4) is also covered by tb_apertif_unb1_correlator_nodes.vhd.
+--
+-- B) Data formats:
-- The beamlet data after the mesh for all Nband*Ndest = 16*8 = 128 PN
-- in the Apertif Correlator together is described by:
--
@@ -69,94 +89,175 @@
-- . pair : Q_interleave
-- . tp_pair : tp MOD Q_interleave
--
+-- C) Maximum and minumum inter channel delay for visibility offload
-- * c_inter_channel_delay
-- With no inter channel output delay, channels are output back-to-back, creating a short full valid burst of 300 visibilities
--- per channel for 64 channels per beamlet. This yields c_nof_complex * cor_out_dat_w * f_clk = 2 * 32 * 200M = 12.8 Gbps in burst.
+-- per channel for 64 channels per beamlet. This yields c_nof_complex * c_cor_out_dat_w * f_clk = 2 * 32 * 200M = 12.8 Gbps in burst.
+-- The link capacity reduces towards the output and is less than:
+-- a) 12.8 Gbps = 200 MHz * 64 bit at node_apertif_unb1_correlator_processing cor_src_out sosi.re/im complex data output
+-- b) 4 Gbps = 125 MHz * 32 bit at node_apertif_unb1_correlator_output sosi.data packet data output
+-- c) 1 Gbps = at 1GbE interface of one node as defined by udp_tx_snk_in_arr with udp_tx_src_out_arr flow control
+-- d) 250 Mbps = 1 Gbps / 4 at 1GbE interface of UniBoard with 8 nodes and one switch with 2 output ports
+-- e) 156.25 Mbps = 20 Gbps / 128 at 2 * 10GbE interface of the Data Writer for 16 UniBoard each with 8 nodes so N_PN = 128 nodes
--
-- * Set c_inter_channel_delay and g_bg_duty_cycle_gap in simulation with 1 PN and 1 GbE link:
-- In the tb with c_nof_tp=2*2=4 there are c_nof_visibilities=10 visibilities, g_nof_beamlets=2 and c_nof_channels=64. Hence the
-- number of output bits per BG sync interval is:
--
--- c_nof_complex * cor_out_dat_w * c_nof_visibilities * c_nof_channels * g_nof_beamlets = 2 * 32 * 10 * 64 * 2
--- = 81920 bits = 1280 complex words
+-- g_nof_beamlets * c_nof_channels * c_nof_visibilities * c_nof_complex * c_cor_out_dat_w = 2 * 64 * 10 * 2 * 32b =
+-- = 81920 bits = 1280 complex words
+-- . c_nof_channels = c_nof_points = 64
+-- . c_nof_blocks_per_sync = c_nof_tsub_per_sync = c_nof_tchan_per_sync * c_nof_channels
+-- . c_nof_frames_per_sync = c_nof_blocks_per_sync
+-- . c_frame_size = g_nof_beamlets
+-- . c_bg_block_period = c_frame_size * Q_interleave + c_bg_duty_cycle_gap
+--
+-- Typical for Apertif correlator the c_bg_block_period = N_clk = 256 dp_clk cycles. To speed up simulation c_bg_duty_cycle_gap
+-- can be set as small as possible:
+--
+-- c_bg_nof_clk_per_sync = c_nof_blocks_per_sync * c_bg_block_period = 640 * (2*2 + 0) = 2560 dp_clk cycles minimal
+--
+-- The minimal BG sync interval in the tb takes c_bg_nof_clk_per_sync = 2560 dp_clk cycles, so g_bg_duty_cycle_gap = 0 can be used,
+-- because 2560 > 1280 at the cor_src_out sosi.re/im interface that can support < 12.8 Gbps. However the data rate at the 1GbE
+-- serial PHY interface must remain < 1Gbps. Per channel there are:
--
--- The BG sync interval in the tb takes c_bg_nof_clk_per_sync = c_nof_blocks_per_sync * c_bg_block_period = 960 * 4 = 3840 clk
--- cycles, so g_bg_duty_cycle_gap=0 can be used, because 3840 > 1280 at the cor_src_out sosi.re/im interface.
--- The data rate at the udp_tx_snk_in_arr(0).re/im interface is 125MHz * 32b = 4 Gbps. However at the 1GbE serial PHY interface
--- the burst data rate must remain < 1Gbps. Per channel there are:
+-- c_vis_header_size = 21 words
+-- c_vis_payload_size = c_nof_visibilities * c_nof_complex = 10 * 2 = 20 words
+-- c_vis_packet_size = c_vis_header_size + c_vis_payload_size = 21 + 20 = 41 words
--
--- gap size = c_network_eth_gap_len * 8b = 96 bits for some idle time between packets
--- header size = 21 * 32b = 672 bits
--- tail size = 1 * 32b = 32 bits
--- payload size = c_nof_complex * cor_out_dat_w * c_nof_visibilities = 2 * 32 * 10 = 640 bits
--- packet size = gap size + header size + payload size + tail size = 96 + 672 + 640 + 32 = 1440 bits
+-- c_vis_nof_packets_per_sync = g_nof_beamlets * c_nof_channels = 2 * 64 = 128
--
--- The transport over 1GbE takes at least 1440 ns @ 1Gbps. One dp_clk cycle is 5 ns (= 1/f_clk = 1/200M), so 1440 / 5 = 288
--- dp_clk cycles. The visibilities arrive in c_nof_visibilities=10 dp_clk cycles in cor_src_out. Therefore:
+-- c_tail_size = 1 word
+-- c_gap_size = c_network_eth_gap_len * 8b / 32b = 3 words for some idle time between packets
+-- c_vis_packet_load = (c_vis_packet_size + c_tail_size + c_gap_size) * 32b = (41 + 1 + 3) * 32b = 45 * 32b = 1440 bits
--
--- c_inter_channel_delay must be > 288 - c_nof_visibilities = 288 - 10 = 278 --> 279
+-- The transport of one packet over 1GbE takes at least c_vis_packet_load / f_link = 1440 ns (f_link = 1Gbps). One dp_clk cycle
+-- is 5 ns (= 1/f_clk = 1/200M), so:
+--
+-- c_vis_packet_nof_dp_clk = c_vis_packet_load * f_clk / f_link = 1440 * 200/1000 = 288 dp_clk cycles
--
--- However with an c_inter_channel_delay the total offload time per sync interval becomes about:
+-- The visibilities per packet arrive in c_nof_visibilities = 10 dp_clk cycles at the cor_src_out sosi.re/im interface. The
+-- c_inter_channel_delay is applied at this interface. Therefore the minimum c_inter_channel_delay is:
--
--- c_inter_channel_delay * c_nof_channels * g_nof_beamlets = 279 * 64 * 2 = 35712 dp_clk cycles
+-- c_inter_channel_delay >= c_vis_packet_nof_dp_clk - c_nof_visibilities = 288 - 10 = 278 --> 279 dp_clk cycles
--
--- This is > BG sync interval of 3840 dp_clk cycles with g_bg_duty_cycle_gap=0. Therefore g_bg_duty_cycle_gap needs to be
+-- With this minimum c_inter_channel_delay the total offload time per sync interval becomes about:
+--
+-- c_vis_nof_packets_per_sync * c_vis_packet_nof_dp_clk = (2 * 64) * 288 = 36864 dp_clk cycles
+--
+-- This is > BG sync interval of 2560 dp_clk cycles with g_bg_duty_cycle_gap = 0. Therefore g_bg_duty_cycle_gap needs to be
-- increased to:
--
--- c_bg_block_period = c_bg_nof_clk_per_sync / c_nof_blocks_per_sync = 35712 / 640 = 55.8 --> 56
--- g_bg_duty_cycle_gap = c_bg_block_period - c_frame_size * Q_interleave = 56 - 2*2 = 52
+-- c_bg_block_period = c_bg_nof_clk_per_sync / c_nof_blocks_per_sync = 36864 / 640 = 57.6 --> 58
+-- g_bg_duty_cycle_gap = c_bg_block_period - c_frame_size * Q_interleave = 58 - 2*2 = 54
+--
+-- Therefore in combination with minimum c_inter_channel_delay = 279 use g_bg_duty_cycle_gap >= 54 to make it work for f_link =
+-- 1Gbps. The maximum g_bg_duty_cycle_gap = N_clk - c_frame_size * Q_interleave = 256 - 88 * 2 = 80 and this is also the typical
+-- g_bg_duty_cycle_gap for the Apertif correlator, because N_clk = 256 is fixed. Hence the typical BG sync interval is:
+--
+-- c_bg_block_period = N_clk = 256
+-- c_bg_nof_clk_per_sync = c_nof_blocks_per_sync * c_bg_block_period = 640 * 256 = 163840 dp_clk cycles typical
+--
+-- The visibility offload must finish within this BG sync interval, hence the maximum number of dp_clk cycles per visibility
+-- packet is:
--
--- Therefore in combination with c_inter_channel_delay=279 use g_bg_duty_cycle_gap >= 52 to make it work for 1GbE @ 1Gbps.
+-- c_vis_packet_offload_nof_dp_clk <= c_bg_nof_clk_per_sync / c_vis_nof_packets_per_sync = 163840 / (2 * 64) = 1280
--
--- To make it work with N_PN = 128 nodes and 2 * 10GbE links to the data writer the c_inter_channel_delay needs to be larger
--- to reduce the rate per 1GbE link to < 20 Gbps / 128 = 156.25 Mbps.
+-- Therefore the maximum c_inter_channel_delay applied at the cor_src_out sosi.re/im interface becomes:
+--
+-- c_inter_channel_delay <= c_vis_packet_offload_nof_dp_clk - c_nof_visibilities = 1280 - 10 = 1270
+--
+-- * Correlator constraint on c_nof_tchan_per_sync:
+-- The correlator.vhd has the constraint that c_nof_tchan_per_sync >= c_nof_mult = c_nof_visibilities / 2**c_nof_pre_mult_folds.
+-- Therefore choose c_nof_tchan_per_sync >= 300 / 2**1 = 150.
--
-- * Set c_inter_channel_delay on hardware with 128 PN and 2 10GbE links:
-- In total there are c_nof_beamlets * c_nof_channels * c_nof_visibilities per t_int = 1.024 sec, so 88 * 64 * 300 * 2 * 32b / 1.024s =
-- 105.6 Mbps on average (8 bit mode with 88 beamlets) or 144 Mbps (6 bit mode with 120 beamlets), which can run over one 1GbE
--- link. The total Apertif correlator output rate for N_PN = 128 nodes to the Apertif data writer is 128 * 105.6M = 13.5168 Gbps
--- respectively 18.432 Gbps, which can run over two 10GbE links.
---
--- We need to set an appropriate number of c_inter_channel_delay cycles for a constant visibility buffer output rate. The
--- correlator outputs c_nof_beamlets * c_nof_channels = 88 * 64 = 5632 respectively 120 * 64 = 7680 blocks of c_nof_visibilities =
--- 300 visibility samples per t_int = N_int / f_sub = 1.024 s = 204.8M dp_clk cycles @ f_clk = 200 MHz. Hence per block there
--- are maximum 36363 or 26666 dp_clk cycles available. The block itself takes c_nof_visibilities = 300 dp_clk cycles. Hence the
--- maximum c_inter_channel_delay becomes 36363 - 300 = 36063 or 26666 - 300 = 26066 dp_clk cycles.
--- The minimum c_inter_channel_delay depends on the capacity of the two 10GbE links. Using the total Apertif correlator output
--- this yields about 13.5168G/20GbE * max(c_inter_channel_delay) = 24373 for 88 beamlets respectively 24022 for 120 beamlets.
---
--- In formula:
+-- link. The total Apertif correlator output rate for N_PN = 128 nodes to the Apertif data writer is 128 * 105.6M = 13.5168 Gbps so
+-- 13.5168/20 = 0.676 utilization, respectively 128 * 144M = 18.432 Gbps so 18.432/20 = 0.922 utilization, which can run over two
+-- 10GbE links.
+--
+-- We need to set an appropriate number of c_inter_channel_delay cycles for a constant visibility buffer output rate.
+-- . The maximum c_inter_channel_delay is determined by the number of cycles per sync interval divided by the number of
+-- visibility packets per sync interval. The correlator outputs c_vis_nof_packets_per_sync = c_nof_beamlets * c_nof_channels
+-- = 88 * 64 = 5632 respectively 120 * 64 = 7680 blocks of c_nof_visibilities = 300 visibility samples per
+-- t_int = N_int / f_sub = 1.024 s = 204.8M dp_clk cycles @ f_clk = 200 MHz. Hence per block there are maximum
+-- 204.8M / 5632 = 36363 or 204.8M / 7680 = 26666 dp_clk cycles available. The block itself takes c_nof_visibilities
+-- = 300 dp_clk cycles. Hence the maximum c_inter_channel_delay becomes 36363 - 300 = 36063 or 26666 - 300 = 26066
+-- dp_clk cycles.
+-- . The minimum c_inter_channel_delay is determined by the available data rate f_link and the processing clock rate of dp_clk.
+-- Important to note is that the burst rate is independent of the packet size, however the minimum inter_channel_delay is
+-- dependent on the packet size, because the inter_channel_delay is applied between packets (so not between samples). Therefore
+-- the minimum c_inter_channel_delay is also determined by the number bits per visibility packet and the number of dp_clk
+-- cycles to process a visibility packet. Per PN the link capacity is f_link = 20G/128 = 156.25 Mbps.
+-- The c_vis_packet_load = (21 + 300*2 + 1 +3) * 32b = 20000 bits per packet. With f_clk = 200 MHz, so 5 ns period, this
+-- corresponds to c_vis_packet_nof_dp_clk = c_vis_packet_load * f_clk / f_link = 20000 * 200M / 156.25 M = 25600
+-- dp_clk cycles. Internally one visibility packet is handled in c_nof_visibilities = 300 dp_clk cycles, so the minimum
+-- c_inter_channel_delay = 25600 - 300 = 25300 dp_clk cycles and is independent of the beamlet bit mode, so the same for
+-- 8 bit and 6 bit.
+-- . As check the ratio between min and max c_inter_channel_delay is (25300 + 300) / (36363 + 300) = 0.698 for 8 bit beamlet
+-- mode and (25300 + 300) / (26666 + 300) = 0.949 for 6 bit beamlet mode.
+--
+-- In formula with HW parameter values:
-- . N_clk = f_clk / f_sub = 200M / 781250 = 256
-- . N_int = c_nof_tsub_per_sync = c_nof_tchan_per_sync*c_nof_points = 12500 * 64 = 800000
-- . c_nof_clk_per_sync = N_int * N_clk = 204.8M
--- . c_nof_visibility_blocks = c_nof_beamlets * c_nof_channels = 5632 for c_nof_beamlets = 88 and c_nof_channels = 64
--- = 7680 for c_nof_beamlets = 120 and c_nof_channels = 64
+-- . c_vis_nof_packets_per_sync = c_nof_beamlets * c_nof_channels = 5632 for c_nof_beamlets = 88 and c_nof_channels = 64
+-- = 7680 for c_nof_beamlets = 120 and c_nof_channels = 64
-- . c_nof_visibilities = 300 = (24*25)/2 for c_nof_tp = 24
--- c_inter_channel_delay = c_nof_clk_per_sync / c_nof_visibility_blocks - c_nof_visibilities = 204.8M / 5632 - 300 = 36063
--- = 204.8M / 7680 - 300 = 26066
---
--- In simulation the N_int needs to be set much smaller than 800000, to shorten the simulation time as much as possible.
--- When c_nof_beamlets is 2 and c_nof_tp = 4 then:
--- . N_clk = 256
--- . N_int = c_nof_tsub_per_sync = c_nof_tchan_per_sync*c_nof_points = 10 * 64
--- . c_nof_clk_per_sync = N_int * N_clk = 163840
--- . c_nof_visibility_blocks = 128 for c_nof_beamlets = 2 and c_nof_channels = 64
--- . c_nof_visibilities = 10 = ( 4* 5)/2 for c_nof_tp = 4
--- c_inter_channel_delay = 163840 / 128 - 10 = 1270 > 0 so this is possible even without throttle BG xon
--- Clearly the c_nof_beamlets has more impact than the c_nof_tp.
---
--- When c_nof_beamlets is 88 and c_nof_tp = 4 then:
--- . N_int = c_nof_tsub_per_sync = c_nof_tchan_per_sync*c_nof_points = 10 * 64
--- . c_nof_clk_per_sync = N_int * N_clk = 163840
--- . c_nof_visibility_blocks = 5632 for c_nof_beamlets = 88 and c_nof_channels = 64
--- . c_nof_visibilities = 10 = ( 4* 5)/2 for c_nof_tp = 4
--- c_inter_channel_delay = 163840 / 5632 - 10 = 19 > 0, so possible.
---
--- If c_nof_tp is increased also then BG xon can be kept active if c_nof_tchan_per_sync is increased also. For
--- c_nof_beamlets = 88 and c_nof_tp = 24 get c_nof_clk_per_sync >= c_nof_visibility_blocks * c_nof_visibilities = 5632 * 300 = 1689600
--- and c_nof_tchan_per_sync = c_nof_clk_per_sync / N_clk / c_nof_points = 1689600 / 256 / 64 = 103, so choose 110. However
--- from the correlator.vhd the constraint is c_nof_tchan_per_sync >= c_nof_mult = c_nof_visibilities / 2**c_nof_pre_mult_folds.
--- Therefore choose c_nof_tchan_per_sync >= 300 / 2**1 = 150.
+-- . f_link = 156.25M
+-- . maximum c_inter_channel_delay = c_nof_clk_per_sync / c_vis_nof_packets_per_sync - c_nof_visibilities
+-- = 204.8M / 5632 - 300 = 36063 = c_hw_inter_channel_delay_max_8bit
+-- = 204.8M / 7680 - 300 = 26366 = c_hw_inter_channel_delay_max_6bit
+-- . minimum c_inter_channel_delay = c_vis_packet_nof_dp_clk - c_nof_visibilities
+-- = (25 + 300*2) * 32b * 200M/156.25M - 300 = 25300 = c_hw_inter_channel_delay_min
+--
+-- In simulation:
+-- The N_int needs to be set much smaller than 800000, to shorten the simulation time as much as possible.
+-- When c_nof_beamlets is 2 and c_nof_tp = 4 then:
+-- . c_bg_block_period = N_clk = 256
+-- . c_nof_visibilities = 10 = (4* 5)/2 for c_nof_tp = 4
+-- . N_int = c_nof_tsub_per_sync = c_nof_tchan_per_sync*c_nof_points = 10 * 64 = 640
+-- . c_nof_clk_per_sync = N_int * c_bg_block_period = 640 * 256 = 163840
+-- . c_vis_nof_packets_per_sync = 128 for c_nof_beamlets = 2 and c_nof_channels = 64
+-- . f_link = 1G, in simulation with one PN use full 1GbE capacity
+-- . maximum c_inter_channel_delay = c_nof_clk_per_sync / c_vis_nof_packets_per_sync - c_nof_visibilities
+-- = 163840 / 128 - 10 = 1270
+-- . minimum c_inter_channel_delay = c_vis_packet_nof_dp_clk - c_nof_visibilities =
+-- = (25 + 10*2) * 32b * 200M/1G - 10 = 278
+-- Trial simulations show:
+-- <= between 230 and 240 it fails because FIFO in node output overflows
+-- >= between 280 and 290 it fails because FIFO in corr_visibility_buffer overflows
+-- Possibly the min value 278 and max value 280 are correct, but the FIFO overflow takes some time to occur,
+-- so the simulation is to short to sharpely reveal the min and max of c_inter_channel_delay.
+--
+-- The fastest simulation is achieved if the c_inter_channel_delay min and max are almost
+-- equal. The maximum c_inter_channel_delay can be reduced by reducing the c_bg_block_period
+-- via g_bg_duty_cycle_gap, this then yields g_bg_duty_cycle_gap = 54 as shown above. To
+-- check with g_bg_duty_cycle_gap = 54 and c_nof_beamlets is 2 and c_nof_tp = 4 then:
+-- . c_bg_block_period = c_frame_size * Q_interleave + c_bg_duty_cycle_gap = 2*2+ 54 = 58
+-- . c_nof_visibilities = 10 = (4* 5)/2 for c_nof_tp = 4
+-- . N_int = c_nof_tsub_per_sync = c_nof_tchan_per_sync*c_nof_points = 10 * 64 = 640
+-- . c_nof_clk_per_sync = N_int * c_bg_block_period = 640 * 58 = 37120
+-- . c_vis_nof_packets_per_sync = 128 for c_nof_beamlets = 2 and c_nof_channels = 64
+-- . f_link = 1G, in simulation with one PN use full 1GbE capacity
+-- . maximum c_inter_channel_delay = c_nof_clk_per_sync / c_vis_nof_packets_per_sync - c_nof_visibilities
+-- = 37120 / 128 - 10 = 280
+-- . minimum c_inter_channel_delay = c_vis_packet_nof_dp_clk - c_nof_visibilities =
+-- = (25 + 10*2) * 32b * 200M/1G - 10 = 278, so choose 279 to fit within min and max
+
+-- When c_nof_beamlets is 2 and c_nof_tp = 24 then:
+-- . c_bg_block_period = N_clk = 256
+-- . c_nof_visibilities = 300 = (24*25)/2 for c_nof_tp = 24
+-- . N_int = c_nof_tsub_per_sync = c_nof_tchan_per_sync*c_nof_points = 160 * 64 = 10240, where correlator constraint 160 > 300/2
+-- . c_nof_clk_per_sync = N_int * c_bg_block_period = 2621440
+-- . c_vis_nof_packets_per_sync = 128 for c_nof_beamlets = 2 and c_nof_channels = 64
+-- . f_link = 1G, in simulation with one PN use full 1GbE capacity
+-- . maximum c_inter_channel_delay = c_nof_clk_per_sync / c_vis_nof_packets_per_sync - c_nof_visibilities
+-- = 2621440 / 128 - 10 = 20470
+-- . minimum c_inter_channel_delay = c_vis_packet_nof_dp_clk - c_nof_visibilities =
+-- = (25 + 300*2) * 32b * 200M/1G - 300 = 3700
--
-- Usage:
-- > as 10
@@ -185,24 +286,25 @@ USE dp_lib.dp_stream_pkg.ALL;
USE diag_lib.diag_pkg.ALL;
USE fft_lib.fft_pkg.ALL;
USE wpfb_lib.wpfb_pkg.ALL;
+USE work.apertif_unb1_correlator_pkg.ALL;
ENTITY tb_node_apertif_unb1_correlator_processing_output IS
GENERIC (
g_tb_end : BOOLEAN := TRUE; -- when TRUE then tb_end ends this simulation, else a higher multi-testbench will end the simulation
- g_tb_index : NATURAL := 0; -- use different index to avoid MM file conflict in multi tb
g_nof_bg_sync : NATURAL := 2; -- [t] number of BG sync intervals, is number ofintegration intervals
-- DUT settings
g_nof_pn : NATURAL := 2; -- number of PN, choose >= 1
g_nof_10G : NATURAL := 2; -- number of 10G input per PN, choose >= 2 and even (to fit Q_interleave=2)
g_nof_beamlets : NATURAL := 2; -- [bu_i], is 88 in 8bit, 120 in 6bit beamlet mode
g_nof_tchan_per_sync : NATURAL := 10; -- [t_c], is 12500, choose > number of taps of WPFB to simulate longer than the FIR impulse response
- g_inter_channel_delay : NATURAL := 279; -- throttle correlator output
+ g_inter_channel_delay : NATURAL := 279; -- throttle correlator output, for expected 279:
g_use_wpfb : BOOLEAN := TRUE;
g_use_prefilter : BOOLEAN := TRUE;
g_use_prefilter_ones : BOOLEAN := FALSE;
-- BG settings
- g_bg_duty_cycle_gap : NATURAL := 56; -- 0 or small for faster simulation, use N_clk - Q_interleave*g_nof_beamlets for HW
+ g_bg_duty_cycle_gap : NATURAL := 54; -- small for faster simulation, use N_clk - Q_interleave*g_nof_beamlets for HW
+ --g_bg_duty_cycle_gap : NATURAL := 256; -- use 256 to default to N_clk - Q_interleave*g_nof_beamlets for HW
g_phasor_ampl : REAL := 0.25; -- range 0:1 where by 1.0 corresponds to maximum +amplitude, use <= 1.0 / 2**c_wpfb_fil_in_backoff_w
g_phasor_freq : REAL := 1.0; -- for N=64 point FFT choose channel in range -32.0 : 0.0 : 31.0
g_phasor_phase : REAL := 0.0 -- [0:2pi>, has nearly no effect on correlator output, because all inputs use the same phasor
@@ -221,6 +323,10 @@ ARCHITECTURE tb OF tb_node_apertif_unb1_correlator_processing_output IS
CONSTANT c_dp_clk_period : TIME := 5 ns;
CONSTANT c_cross_clock_domain_latency : NATURAL := 20;
+ CONSTANT f_link : REAL := 1000.0; -- 1 Gbps
+ --CONSTANT f_link : REAL := 156.25;
+ CONSTANT f_dp_clk : REAL := 200.0; -- 200 MHz
+
-- UniBoard
CONSTANT c_unb_nr : NATURAL := 14; -- Uniboard 0:15
CONSTANT c_fpga_nr : NATURAL := 5; -- FPGA 0:7 on a UniBoard, FPGA 0:3 = FN 0:3, FPGA 4:7 = BN 0:3
@@ -269,32 +375,41 @@ ARCHITECTURE tb OF tb_node_apertif_unb1_correlator_processing_output IS
CONSTANT c_bg_nof_clk_per_sync : NATURAL := c_nof_blocks_per_sync * c_bg_block_period;
-- . correlator DUT input
- CONSTANT c_nof_tp : NATURAL := g_nof_pn * g_nof_10G; -- number of telescope paths
+ CONSTANT c_nof_tp : NATURAL := g_nof_pn * g_nof_10G; -- number of telescope paths
- CONSTANT c_nof_visibilities : NATURAL := c_nof_tp * (c_nof_tp+1) / 2;
- CONSTANT c_accu_nof_samples : NATURAL := c_nof_tsub_per_sync; -- [ti]
- CONSTANT c_inter_channel_delay : NATURAL := g_inter_channel_delay;
+ CONSTANT c_nof_visibilities : NATURAL := c_nof_tp * (c_nof_tp+1) / 2;
+ CONSTANT c_accu_nof_samples : NATURAL := c_nof_tsub_per_sync; -- [ti]
+ CONSTANT c_inter_channel_delay : NATURAL := g_inter_channel_delay;
- CONSTANT c_in_dat_w : NATURAL := c_wpfb_apertif_channels.fil_in_dat_w; -- = 8 bit
- CONSTANT c_in_complex_w : NATURAL := c_nof_complex * c_in_dat_w; -- = 16 bit, complex Im & Re
- CONSTANT c_cor_out_dat_w : NATURAL := 32;
+ CONSTANT c_in_dat_w : NATURAL := c_wpfb_apertif_channels.fil_in_dat_w; -- = 8 bit
+ CONSTANT c_in_complex_w : NATURAL := c_nof_complex * c_in_dat_w; -- = 16 bit, complex Im & Re
+ CONSTANT c_cor_out_dat_w : NATURAL := 32;
- -- . DUT output DB
- CONSTANT c_vis_header_size : NATURAL := 21; -- (pad(2) + eth(14) + ip(20) + udp(8) + app_id(16) + app_flags(24)) / 4 = 84 bytes / 4 = 21 words
- CONSTANT c_vis_payload_size : NATURAL := c_nof_complex*c_nof_visibilities;
- CONSTANT c_vis_packet_size : NATURAL := c_vis_header_size + c_vis_payload_size;
+ -- . 1GbE output
+ CONSTANT c_packet_info : t_apertif_unb1_correlator_packet_info := func_apertif_unb1_correlator_packet_info(c_nof_visibilities, f_link, f_dp_clk);
+ CONSTANT c_inter_channel_delay_min : INTEGER := func_apertif_unb1_correlator_inter_channel_delay_min(c_packet_info, c_nof_visibilities);
+ CONSTANT c_inter_channel_delay_max : INTEGER := func_apertif_unb1_correlator_inter_channel_delay_max(c_bg_block_period, c_bg_nof_blocks_per_sync, g_nof_beamlets, c_nof_channels, c_nof_visibilities);
+
+ CONSTANT c_vis_header_size : NATURAL := c_packet_info.vis_header_size; -- = (pad(2) + eth(14) + ip(20) + udp(8) + app_id(16) + app_flags(24)) / 4 = 84 bytes / 4 = 21 words
+ CONSTANT c_vis_payload_size : NATURAL := c_packet_info.vis_payload_size; -- = c_nof_complex*c_nof_visibilities;
+ CONSTANT c_vis_packet_size : NATURAL := c_packet_info.vis_packet_size; -- = c_vis_header_size + c_vis_payload_size;
CONSTANT c_vis_nof_packets_per_sync : NATURAL := g_nof_beamlets * c_nof_channels;
CONSTANT c_vis_nof_data_per_sync : NATURAL := c_vis_nof_packets_per_sync * c_vis_packet_size;
- --CONSTANT c_db_sync_delay : NATURAL := c_vis_nof_data_per_sync - c_vis_packet_size; -- at last packet in sync interval
- CONSTANT c_db_sync_delay : NATURAL := 0;
- -- . 1GbE output
CONSTANT c_eth_check_nof_packets : NATURAL := c_nof_cor_sync * c_nof_channels;
- CONSTANT c_eth_header_size : NATURAL := 21; -- (pad(2) + eth(14) + ip(20) + udp(8) + app(16+24))/4 = 84 / 4
- CONSTANT c_udp_payload_size : NATURAL := c_nof_complex * c_nof_visibilities;
- CONSTANT c_eth_packet_size : NATURAL := c_eth_header_size + c_udp_payload_size;
-
+ CONSTANT c_eth_runtime_timeout : TIME := g_nof_bg_sync * c_bg_nof_clk_per_sync * c_dp_clk_period * 2; -- factor 2 margin
+ SIGNAL dbg_c_eth_runtime_timeout : TIME := c_eth_runtime_timeout;
+ -- . DUT output DB
+ --CONSTANT c_db_sync_delay : NATURAL := c_vis_nof_data_per_sync - c_vis_packet_size; -- at last packet in sync interval
+ CONSTANT c_db_sync_delay : NATURAL := 0;
+
+ -- Show parameter values for HW generics in Wave window for debugging purposes
+ CONSTANT c_hw_packet_info : t_apertif_unb1_correlator_packet_info := func_apertif_unb1_correlator_packet_info(300, 156.25, 200.0);
+ CONSTANT c_hw_inter_channel_delay_min : INTEGER := func_apertif_unb1_correlator_inter_channel_delay_min(c_hw_packet_info, 300);
+ CONSTANT c_hw_inter_channel_delay_max_8bit : INTEGER := func_apertif_unb1_correlator_inter_channel_delay_max(256, 800000, 88, 64, 300);
+ CONSTANT c_hw_inter_channel_delay_max_6bit : INTEGER := func_apertif_unb1_correlator_inter_channel_delay_max(256, 800000, 120, 64, 300);
+
-- Phasor: exp(j*angle) = cos(angle) + j*sin(angle)
-- A complex FFT of N points has N bins or channels: ch = -N/2:0:N/2-1.
-- To create an FFT input phasor with frequency in the middle of a channel use FREQ = ch.
@@ -372,6 +487,18 @@ ARCHITECTURE tb OF tb_node_apertif_unb1_correlator_processing_output IS
SIGNAL dp_rst : STD_LOGIC := '1';
SIGNAL verify_activated : STD_LOGIC := '0';
+ -- Show parameter values for Tb generics in Wave window for debugging purposes
+ SIGNAL dbg_c_packet_info : t_apertif_unb1_correlator_packet_info := func_apertif_unb1_correlator_packet_info(c_nof_visibilities, f_link, f_dp_clk);
+ SIGNAL dbg_c_inter_channel_delay : INTEGER := c_inter_channel_delay;
+ SIGNAL dbg_c_inter_channel_delay_min : INTEGER := c_inter_channel_delay_min;
+ SIGNAL dbg_c_inter_channel_delay_max : INTEGER := c_inter_channel_delay_max;
+
+ -- Show parameter values for HW generics in Wave window for debugging purposes
+ SIGNAL dbg_c_hw_packet_info : t_apertif_unb1_correlator_packet_info := c_hw_packet_info;
+ SIGNAL dbg_c_hw_inter_channel_delay_min : INTEGER := c_hw_inter_channel_delay_min;
+ SIGNAL dbg_c_hw_inter_channel_delay_max_6bit : INTEGER := c_hw_inter_channel_delay_max_6bit;
+ SIGNAL dbg_c_hw_inter_channel_delay_max_8bit : INTEGER := c_hw_inter_channel_delay_max_8bit;
+
SIGNAL bg_start : STD_LOGIC;
SIGNAL bg_sync_cnt : NATURAL := 0;
SIGNAL cor_sync_cnt : NATURAL;
@@ -418,6 +545,24 @@ BEGIN
dp_rst <= '1', '0' AFTER c_dp_clk_period*7;
mm_rst <= '1', '0' AFTER c_mm_clk_period*7;
+ p_check_inter_channel_delay : PROCESS
+ VARIABLE v_bool : BOOLEAN := TRUE;
+ --VARIABLE v_bool : BOOLEAN := FALSE;
+ BEGIN
+ WAIT FOR 1 us;
+ IF v_bool THEN
+ v_bool := func_apertif_unb1_correlator_verify_and_log_output_rate(c_inter_channel_delay,
+ c_bg_block_period,
+ c_nof_blocks_per_sync,
+ c_nof_beamlets,
+ c_nof_channels,
+ c_nof_visibilities,
+ f_link,
+ f_dp_clk);
+ END IF;
+ WAIT;
+ END PROCESS;
+
p_mm_stimuli : PROCESS
CONSTANT c_tp_force_zero : BOOLEAN := FALSE; -- TRUE to understand corr_folder order of [tp_pair][pair], else FALSE
VARIABLE c_tp_sel : NATURAL := 0; -- selected TP for force data parallel
@@ -636,9 +781,9 @@ BEGIN
u_eth_statistics : ENTITY eth_lib.eth_statistics
GENERIC MAP (
g_runtime_nof_packets => c_eth_check_nof_packets,
- g_runtime_timeout => 1000 us,
+ g_runtime_timeout => c_eth_runtime_timeout,
g_check_nof_valid => TRUE,
- g_check_nof_valid_ref => c_eth_check_nof_packets*c_eth_packet_size
+ g_check_nof_valid_ref => c_eth_check_nof_packets*c_vis_packet_size
)
PORT MAP (
eth_serial_in => eth_sgout,
@@ -706,7 +851,7 @@ BEGIN
VARIABLE v_rd_vis_im : INTEGER;
VARIABLE v_rd_timestamp : STD_LOGIC_VECTOR(63 DOWNTO 0);
VARIABLE v_exp_beamlet : NATURAL; -- interleaved beamlet index
- VARIABLE v_exp_vis_channel : NATURAL; -- channel index in visibility offload packet
+ VARIABLE v_exp_channel : NATURAL; -- channel index in visibility offload packet
VARIABLE v_exp_vis_re : INTEGER;
VARIABLE v_exp_vis_im : INTEGER;
VARIABLE v_exp_timestamp : STD_LOGIC_VECTOR(63 DOWNTO 0);
@@ -755,8 +900,8 @@ BEGIN
v_exp_beamlet := v_exp_beamlet + vB * Q_interleave; -- offset serial beamlets
FOR vC IN 0 TO c_nof_channels-1 LOOP
v_addr := ((vB * c_nof_channels) + vC) * c_vis_packet_size;
- v_exp_vis_channel := flip(vC, c_nof_channels_w);
- v_exp_vis_channel := fft_shift(v_exp_vis_channel, c_nof_channels_w);
+ v_exp_channel := flip(vC, c_nof_channels_w);
+ v_exp_channel := fft_shift(v_exp_channel, c_nof_channels_w);
-----------------------------------------------------------------------
-- Verify header fields
-- 0 gap(16) + ETH dst mac hi(16)
@@ -807,8 +952,8 @@ BEGIN
proc_mem_mm_bus_rd(v_addr + 12, mm_clk, ram_diag_data_buffer_output_miso, ram_diag_data_buffer_output_mosi);
proc_mem_mm_bus_rd_latency(c_mem_reg_rd_latency, mm_clk);
v_rd_channel := TO_UINT(ram_diag_data_buffer_output_miso.rddata(31 DOWNTO 16));
- ASSERT v_rd_beamlet = v_exp_beamlet AND v_rd_channel = v_exp_vis_channel REPORT "Wrong beamlet, channel index in packet header: " &
- "(" & int_to_str(v_exp_beamlet) & ", " & int_to_str(v_exp_vis_channel) & ") /= " &
+ ASSERT v_rd_beamlet = v_exp_beamlet AND v_rd_channel = v_exp_channel REPORT "Wrong beamlet, channel index in packet header: " &
+ "(" & int_to_str(v_exp_beamlet) & ", " & int_to_str(v_exp_channel) & ") /= " &
"(" & int_to_str(v_rd_beamlet) & ", " & int_to_str(v_rd_channel) & ")" SEVERITY ERROR;
-- . timestamp (= bsn)
@@ -866,10 +1011,10 @@ BEGIN
END IF;
-- . verify
ASSERT v_rd_vis_re = v_exp_vis_re REPORT "Wrong real visibility for beamlet, channel index in packet payload " &
- "(" & int_to_str(v_exp_beamlet) & ", " & int_to_str(v_exp_vis_channel) & ") : " &
+ "(" & int_to_str(v_exp_beamlet) & ", " & int_to_str(v_exp_channel) & ") : " &
int_to_str(v_rd_vis_re) & " /= " & int_to_str(v_exp_vis_re) SEVERITY ERROR;
ASSERT v_rd_vis_im = v_exp_vis_im REPORT "Wrong imag visibility for beamlet, channel index in packet payload " &
- "(" & int_to_str(v_exp_beamlet) & ", " & int_to_str(v_exp_vis_channel) & ") : " &
+ "(" & int_to_str(v_exp_beamlet) & ", " & int_to_str(v_exp_channel) & ") : " &
int_to_str(v_rd_vis_im) & " /= " & int_to_str(v_exp_vis_im) SEVERITY ERROR;
END LOOP;
END LOOP;