5 Releases 5 Tags

RSS Feed

• v2.0.0-fft2048 2401f5f89e

  Compare

  Stable

  mirror released this 2026-04-16 17:25:05 +00:00 | 37 commits to main since this release

  📅 Originally published on GitHub: Thu, 16 Apr 2026 17:25:56 GMT
  🏷️ Git tag created: Thu, 16 Apr 2026 17:25:05 GMT

  AERIS-10 2048-pt FFT Upgrade + Full Timing Closure

  Complete cross-layer upgrade from 1024-pt to 2048-pt FFT with decimation=4, producing 512 output range bins at 6m spacing (was 64 bins at 24m). All 5 FPGA clock domains timing-closed on xc7a50tftg256-2 after 3 build iterations.

  Timing (Build 19)

  Clock	Period	WNS	Failing Endpoints	Previous (v1.1.0)
  clk_mmcm_out0 (400 MHz)	2.5 ns	+0.068 ns	0	+0.339 ns
  clk_100m (100 MHz)	10.0 ns	+0.156 ns	0	+0.080 ns
  clk_120m_dac (120 MHz)	8.333 ns	+0.627 ns	0	—
  ft_clkout (60 MHz)	16.667 ns	+9.887 ns	0	—
  adc_dco_p (400 MHz)	2.5 ns	+0.920 ns	0	—
  Hold (WHS)	—	+0.046 ns	0	+0.056 ns
  Pulse width (WPWS)	—	+0.361 ns	0	+0.361 ns
  All constraints met		Yes		Yes

  Utilization (xc7a50tftg256-2)

  Resource	Used	Available	Util%	Previous (v1.1.0)
  LUTs	21,735	32,600	66.67%	10,252 (31.4%)
  Flip-Flops	14,579	65,200	22.36%	12,820 (19.7%)
  BRAM	55.5	75	74.00%	17.5 (23.3%)
  DSP48E1	112	120	93.33%	112 (93.3%)
  F7 Muxes	7,849	16,300	48.15%	—
  F8 Muxes	3,336	8,150	40.93%	—

  What Changed (from v1.1.0-agc)

  2048-pt FFT / 512-Bin DSP Pipeline

  • fft_engine.v: 2048-pt FFT with 11 butterfly stages (was 1024/10)
  • range_bin_decimator.v: 4x decimation (was 16x), 512 output bins at 6m spacing
  • doppler_processor.v: 16384-deep Doppler memory (was 2048), 14-bit addressing
  • chirp_memory_loader_param.v: 2-segment loader (was 4), 2048-sample arrays
  • matched_filter_multi_segment.v: 11-bit sample addressing, 2-segment processing
  • radar_receiver_final.v: 9-bit range bin width, sample_addr truncation fix
  • usb_data_interface_ft2232h.v: Bulk per-frame format with Manhattan magnitude
  • radar_params.vh: Single source of truth via include in all RTL modules

  FPGA Timing Closure (Builds 17–19)

  • CFAR pipeline fix (cfar_ca.v): Pre-registered col_buf reads break 15-level mux tree critical path. clk_100m WNS: -0.331ns → +0.156ns
  • CIC reset path fix (cic_decimator_4x_enhanced.v, ddc_400m.v): Eliminated LUT1 inverter on reset fan-out to 8 DSP48E1 RSTB pins. Wired pre-registered reset_400m from DDC. clk_mmcm_out0 WNS: -0.074ns → +0.068ns

  BRAM Inference Fixes (Build 16 → 17)

  • mti_canceller.v: prev_i/q arrays converted from async to sync reset → BRAM inference (saved ~16K fabric FFs)
  • matched_filter_multi_segment.v: overlap_cache sync-only write block → BRAM inference

  Cross-Layer Updates

  • Python GUI: radar_protocol.py bulk frame parser for 512-bin arrays, 6m range scale, 6-bit stream control. Both tkinter and V7 PyQt6 dashboards updated
  • MCU firmware: RadarSettings.cpp max_distance 1536→3072m, map_size updated
  • Golden references: All .hex, .csv, .npy co-sim data regenerated for 2048/512 config
  • Stale cleanup: Deleted dead test classes, stale chirp segment files, dead utility functions

  Test Suite

  Layer	Tests	Status
  Python (GUI)	172 (tkinter + PyQt6)	Pass
  MCU (C/C++)	22	Pass
  FPGA (iverilog)	30	Pass
  Cross-layer	61	Pass
  Total	285	All pass, 0 skip

  Assets

  • radar_system_top_50t.bit — Production bitstream (2,140 KB, program via Vivado/xsdb)
  • 02_timing_summary.rpt — Post-route timing report
  • 04_utilization.rpt — Post-route utilization report

  Build Environment

  • Vivado 2025.2 (lin64)
  • Device: xc7a50tftg256-2 (speed grade -2, PRODUCTION)
  • Top module: radar_system_top_50t
  • Synth: 214s, Impl: 1631s, Bit: 38s
  • Build directives: ExtraNetDelay_high place, 3x AggressiveExplore phys_opt

  Downloads

  • 02_timing_summary.rpt
    106 KiB

    2026-04-19 11:27:58 +00:00
  • 04_utilization.rpt
    12 KiB

    2026-04-19 11:27:58 +00:00
  • radar_system_top_50t.bit
    2.1 MiB

    2026-04-19 11:28:00 +00:00
  • Source Code (ZIP)
  • Source Code (TAR.GZ)