7742b517b6
fix(fpga): implement 5 P0 invariant fixes with adversarial testbenches
...
Fixes 5 critical cross-layer invariant violations found during
system-level analysis. Each fix has a dedicated adversarial testbench
that actively tries to break the fix under race conditions, reset
mid-operation, overflow, and pathological input patterns.
RTL fixes:
- Fix #1 : Replace flawed cdc_adc_to_processing with Gray-coded async
FIFO (cdc_async_fifo) for DDC 400->100 MHz CDC path. Pre-fetch
show-ahead architecture with CDC-safe registered reads.
- Fix #2 : XOR toggle detection for mc_new_chirp in matched filter
(cross-clock-domain safe vs level-sensitive).
- Fix #3 : ST_WAIT_LISTEN state with configurable listen_delay to
prevent matched filter re-trigger during chirp dead time.
- Fix #4 : Overlap-save output trim in matched filter to suppress
circular convolution artifacts at segment boundaries.
- Fix #7 : Falling-edge frame_complete pulse in doppler_processor
(was stuck high, causing continuous AGC resets).
RTL cleanup:
- Refactor CDC synchronizer arrays from memory arrays to scalar regs
for explicit ASYNC_REG flop naming and synthesis constraint clarity.
Testbenches (70 checks total, all passing):
- tb_p0_async_fifo.v: 20 checks (fill, overflow, reset, streaming,
show-ahead capacity, pathological data patterns)
- tb_p0_mf_adversarial.v: 33 checks (toggle detection, listen state,
overlap trim, rapid chirp sequences, reset recovery)
- tb_p0_frame_pulse.v: 17 checks (pulse width, idle behavior,
processing duration sweep, regression vs old stuck-high bug)
Regression: 24/24 pass (--quick), 57/57 existing CDC tests pass.
Golden references updated for doppler output timing change.
2026-04-17 01:53:06 +05:45
063fa081fe
fix: FPGA timing margins (WNS +0.002→+0.080ns) + 11 bug fixes from code review
...
FPGA timing (400MHz domain WNS: +0.339ns, was +0.002ns):
- DONT_TOUCH on BUFG to prevent AggressiveExplore cascade replication
- NCO→mixer pipeline registers break critical 1.5ns route
- Clock uncertainty reduced 200ps→100ps (adequate guardband)
- Updated golden/cosim references for +1 cycle pipeline latency
STM32 bug fixes:
- Guard uint32_t underflow in processStartFlag (length<4)
- Replace unbounded strcat in getSystemStatusForGUI with snprintf
- Early-return error masking in checkSystemHealth
- Add HAL_Delay in emergency blink loop
GUI bug fixes:
- Remove 0x03 from _HARDWARE_ONLY_OPCODES (was in both sets)
- Wire real error count in V7 diagnostics panel
- Fix _stop_demo showing 'Live' label during replay mode
FPGA comment fixes + CI: add test_v7.py to pytest command
Vivado build 50t passed: 0 failing endpoints, WHS=+0.056ns
2026-04-14 00:08:26 +05:45
1e284767cd
fix(test,docs): remove dead xfft_32 files, update test infra for dual-16 FFT, add regression guide
...
- Remove xfft_32.v, tb_xfft_32.v, and fft_twiddle_32.mem (dead code
since PR #33 moved Doppler to dual 16-pt FFT architecture)
- Update run_regression.sh: xfft_16 in PROD_RTL, remove xfft_32 from
EXTRA_RTL and all compile commands
- Update tb_fft_engine.v to test with N=16 / fft_twiddle_16.mem
- Update validate_mem_files.py: validate fft_twiddle_16.mem instead of 32
- Update testbenches and golden data from main_cleanup branch to match
dual-16 architecture (tb_doppler_cosim, tb_doppler_realdata,
tb_fullchain_realdata, tb_fullchain_mti_cfar_realdata, tb_system_e2e,
radar_receiver_final, golden_doppler.mem)
- Update CONTRIBUTING.md with full regression test instructions covering
FPGA, MCU, GUI, co-simulation, and formal verification
Regression: 23/23 FPGA, 20/20 MCU, 57/58 GUI, 56/56 mem validation,
all co-sim scenarios PASS.
2026-04-07 02:51:48 +03:00
4985eccbae
Wire self-test results (0x31) to USB status readback path, add fpga_self_test to regression
...
- usb_data_interface.v: Add 3 self-test status inputs, expand status packet
from 7 words (header + 5 data + footer) to 8 words (header + 6 data + footer).
New status_words[5] carries {busy, detail[7:0], flags[4:0]}.
- radar_system_top.v: Wire self_test_flags_latched, self_test_detail_latched,
self_test_busy to usb_data_interface ports. Add opcode 0x31 as status
readback alias so host can read self-test results.
- tb_usb_data_interface.v: Add self-test port connections, verify word 5 in
Group 16, add Group 18 (busy flag + partial failure variant). 81 checks pass.
- run_regression.sh: Add fpga_self_test.v to PROD_RTL lint list and system-
level compile lists. Add tb_fpga_self_test as Phase 1 unit test.
- 24/24 regression tests pass, lint clean (0 errors, 4 advisory warnings).
2026-03-20 20:03:11 +02:00
e93bc33c6c
Production fixes 1-7: detection bugs, cfar→threshold rename, digital gain control, Doppler mismatch protection, decimator watchdog, bypass_mode dead code removal, range-mode register (21/21 regression PASS)
...
Fix 1: Combinational magnitude + non-sticky detection flag (tb: 23/23)
Fix 2: Rename all cfar_* signals to detect_*/threshold_* (honest naming)
Fix 3: New rx_gain_control.v between DDC and FFT, opcode 0x16 (tb: 33/33)
Fix 4: Clamp host_chirps_per_elev to DOPPLER_FFT_SIZE, error flag (E2E: 54/54)
Fix 5: Decimator watchdog timeout, 256-cycle limit (tb: 63/63)
Fix 6: Remove bypass_mode dead code from ddc_400m.v (DDC tb: 21/21)
Fix 7: Range-mode register 0x20 with status readback (USB tb: 77/77)
2026-03-20 04:38:35 +02:00
c6103b37de
Gap 7 MMCM jitter cleaner + CIC comb CREG pipeline + XDC clock-name fix
...
MMCM (Gap 7):
- Add adc_clk_mmcm.v: MMCME2_ADV wrapper (VCO=800MHz, CLKOUT0=400MHz)
- Modify ad9484_interface_400m.v: replace BUFG with MMCM path, gate reset on mmcm_locked
- Add adc_clk_mmcm.xdc: CDC false paths for clk_mmcm_out0 <-> clk_100m
XDC Fix (Build 19 WNS=-0.011 root cause):
- Remove conflicting create_generated_clock -name clk_400m_mmcm
- Replace all clk_400m_mmcm references with Vivado auto-generated clk_mmcm_out0
- CDC false paths now correctly apply to actual timing paths
CIC CREG Pipeline (Build 18 critical path fix):
- Explicit DSP48E1 for comb[0] with CREG=1/AREG=1/BREG=1/PREG=1
- Absorbs integrator_sampled_comb fabric FDRE into DSP48 C-port register
- Eliminates 0.643ns fabric->DSP routing delay (Build 18 tightest path)
- +1 cycle comb latency via data_valid_comb_0_out pipeline
- Move shared register declarations above ifndef SIMULATION (iverilog fix)
- Update golden data for +1 cycle CIC pipeline shift
Build scripts: build19_mmcm.tcl, build20_mmcm_creg.tcl
Regression: 18/18 FPGA pass, 20/20 MCU pass
Build 20 launched on remote Vivado (pending results)
2026-03-19 22:59:46 +02:00
94ffdb8f77
Add Phase 0 Vivado-style lint to regression runner, update golden data
...
Adds two-layer lint pass (iverilog -Wall + custom static checks) that
catches part-select OOB errors and case-without-default warnings before
pushing to remote Vivado. Catches the exact Synth 8-524 class error that
broke Build 18 initial attempt. Lint errors abort regression; warnings
are advisory. Regenerated golden data for BRAM-migrated matched filter.
2026-03-19 21:19:07 +02:00
ed6f79c6d3
FIR DSP48 pipelining (BREG+MREG) + matched filter BRAM migration with overlap cache
...
FIR: Add coeff_reg/mult_reg pipeline stages to fix 68 DPIP-1 + 35 DPOP-2
DRC warnings. Valid pipeline widened 7→9 bits (+2 cycle latency).
Matched filter: Migrate input_buffer_i/q from register arrays to BRAM
(~33K FF savings). Overlap-save uses register cache captured during
ST_PROCESSING to avoid BRAM read/write conflicts during overlap copy.
New ST_OVERLAP_COPY state writes cached tail samples back sequentially.
Both changes pass 18/18 FPGA regression. Golden data regenerated for
+2 FIR latency baseline.
2026-03-19 20:39:01 +02:00
47606a4459
Rewrite integration testbench with golden self-reference comparison + physics bounds checks
...
Replace smoke-test integration TB (10 liveness checks) with golden
comparison architecture (18 checks). Two compile-time modes:
- GOLDEN_GENERATE: dumps 2048 Doppler outputs to golden_doppler.mem
- Default: loads golden, compares within ±2 LSB tolerance per channel
New checks: G1 golden comparison (2048/2048 exact match verified),
B1a/b DDC energy bounds, B2a/b Doppler per-bin energy, B3 exact count,
B4 full 64x32 bin coverage, B5 no duplicate indices.
Fault injection verified: zeroing FIR coeff[15] causes 2048/2048
golden mismatches (max err 1234/1443 LSB), confirming regression guard.
2026-03-17 20:56:28 +02:00
Jason