FlintyLemming/PLFM_RADAR
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: FPGA timing margins (WNS +0.002→+0.080ns) + 11 bug fixes from code review

Browse Source 
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
This commit is contained in:
Jason
2026-04-14 00:08:26 +05:45
parent b4d1869582
commit 063fa081fe
12 changed files with 4323 additions and 4262 deletions
Download Patch File Download Diff File Expand all files Collapse all files
9_Firmware/9_2_FPGA/ddc_400m.v
+46 -16
View File
@@ -102,14 +102,19 @@ wire signed [17:0] debug_mixed_q_trunc;
reg [7:0] signal_power_i, signal_power_q;
// Internal mixing signals
// DSP48E1 with AREG=1, BREG=1, MREG=1, PREG=1 handles all internal pipelining
// Latency: 4 cycles (1 for AREG/BREG, 1 for MREG, 1 for PREG, 1 for post-DSP retiming)
// Pipeline: NCO fabric reg (1) + DSP48E1 AREG/BREG (1) + MREG (1) + PREG (1) + retiming (1) = 5 cycles
// The NCO fabric pipeline register was added to break the long NCO→DSP B-port route
// (1.505ns routing in Build 26, WNS=+0.002ns). With BREG=1 still active inside the DSP,
// total latency increases by 1 cycle (2.5ns at 400MHz — negligible for radar).
wire signed [MIXER_WIDTH-1:0] adc_signed_w;
reg signed [MIXER_WIDTH + NCO_WIDTH -1:0] mixed_i, mixed_q;
reg mixed_valid;
reg mixer_overflow_i, mixer_overflow_q;
// Pipeline valid tracking: 4-stage shift register (3 for DSP48E1 + 1 for post-DSP retiming)
reg [3:0] dsp_valid_pipe;
// Pipeline valid tracking: 5-stage shift register (1 NCO pipe + 3 DSP48E1 + 1 retiming)
reg [4:0] dsp_valid_pipe;
// NCO→DSP pipeline registers — breaks the long NCO sin/cos → DSP48E1 B-port route
// DONT_TOUCH prevents Vivado from absorbing these into the DSP or optimizing away
(* DONT_TOUCH = "TRUE" *) reg signed [15:0] cos_nco_pipe, sin_nco_pipe;
// Post-DSP retiming registers — breaks DSP48E1 CLK→P to fabric timing path
// This extra pipeline stage absorbs the 1.866ns DSP output prop delay + routing,
// ensuring WNS > 0 at 400 MHz regardless of placement seed
@@ -210,11 +215,11 @@ nco_400m_enhanced nco_core (
//
// Architecture:
// ADC data → sign-extend to 18b → DSP48E1 A-port (AREG=1 pipelines it)
// NCO cos/sin → sign-extend to 18b → DSP48E1 B-port (BREG=1 pipelines it)
// NCO cos/sin → fabric pipeline reg → DSP48E1 B-port (BREG=1 pipelines it)
// Multiply result captured by MREG=1, then output registered by PREG=1
// force_saturation override applied AFTER DSP48E1 output (not on input path)
//
// Latency: 3 clock cycles (AREG/BREG + MREG + PREG)
// Latency: 4 clock cycles (1 NCO pipe + 1 AREG/BREG + 1 MREG + 1 PREG) + 1 retiming = 5 total
// PREG=1 absorbs DSP48E1 CLK→P delay internally, preventing fabric timing violations
// In simulation (Icarus), uses behavioral equivalent since DSP48E1 is Xilinx-only
// ============================================================================
@@ -223,24 +228,35 @@ nco_400m_enhanced nco_core (
assign adc_signed_w = {1'b0, adc_data, {(MIXER_WIDTH-ADC_WIDTH-1){1'b0}}} -
{1'b0, {ADC_WIDTH{1'b1}}, {(MIXER_WIDTH-ADC_WIDTH-1){1'b0}}} / 2;
// Valid pipeline: 4-stage shift register (3 for DSP48E1 AREG+MREG+PREG + 1 for retiming)
// Valid pipeline: 5-stage shift register (1 NCO pipe + 3 DSP48E1 AREG+MREG+PREG + 1 retiming)
always @(posedge clk_400m or negedge reset_n_400m) begin
if (!reset_n_400m) begin
dsp_valid_pipe <= 4'b0000;
dsp_valid_pipe <= 5'b00000;
end else begin
dsp_valid_pipe <= {dsp_valid_pipe[2:0], (nco_ready && adc_data_valid_i && adc_data_valid_q)};
dsp_valid_pipe <= {dsp_valid_pipe[3:0], (nco_ready && adc_data_valid_i && adc_data_valid_q)};
end
end
`ifdef SIMULATION
// ---- Behavioral model for Icarus Verilog simulation ----
// Mimics DSP48E1 with AREG=1, BREG=1, MREG=1, PREG=1 (3-cycle latency)
// Mimics NCO pipeline + DSP48E1 with AREG=1, BREG=1, MREG=1, PREG=1 (4-cycle DSP + 1 NCO pipe)
reg signed [MIXER_WIDTH-1:0] adc_signed_reg; // Models AREG
reg signed [15:0] cos_pipe_reg, sin_pipe_reg; // Models BREG
reg signed [MIXER_WIDTH+NCO_WIDTH-1:0] mult_i_internal, mult_q_internal; // Models MREG
reg signed [MIXER_WIDTH+NCO_WIDTH-1:0] mult_i_reg, mult_q_reg; // Models PREG
// Stage 1: AREG/BREG equivalent
// Stage 0: NCO pipeline — breaks long NCO→DSP route (matches synthesis fabric registers)
always @(posedge clk_400m or negedge reset_n_400m) begin
if (!reset_n_400m) begin
cos_nco_pipe <= 0;
sin_nco_pipe <= 0;
end else begin
cos_nco_pipe <= cos_out;
sin_nco_pipe <= sin_out;
end
end
// Stage 1: AREG/BREG equivalent (uses pipelined NCO outputs)
always @(posedge clk_400m or negedge reset_n_400m) begin
if (!reset_n_400m) begin
adc_signed_reg <= 0;
@@ -248,8 +264,8 @@ always @(posedge clk_400m or negedge reset_n_400m) begin
sin_pipe_reg <= 0;
end else begin
adc_signed_reg <= adc_signed_w;
cos_pipe_reg <= cos_out;
sin_pipe_reg <= sin_out;
cos_pipe_reg <= cos_nco_pipe;
sin_pipe_reg <= sin_nco_pipe;
end
end
@@ -291,6 +307,20 @@ end
// This guarantees AREG/BREG/MREG are used, achieving timing closure at 400 MHz
wire [47:0] dsp_p_i, dsp_p_q;
// NCO pipeline stage — breaks the long NCO sin/cos → DSP48E1 B-port route
// (1.505ns routing observed in Build 26). These fabric registers are placed
// near the DSP by the placer, splitting the route into two shorter segments.
// DONT_TOUCH on the reg declaration (above) prevents absorption/retiming.
always @(posedge clk_400m or negedge reset_n_400m) begin
if (!reset_n_400m) begin
cos_nco_pipe <= 0;
sin_nco_pipe <= 0;
end else begin
cos_nco_pipe <= cos_out;
sin_nco_pipe <= sin_out;
end
end
// DSP48E1 for I-channel mixer (adc_signed * cos_out)
DSP48E1 #(
// Feature control attributes
@@ -350,7 +380,7 @@ DSP48E1 #(
.CEINMODE(1'b0),
// Data ports
.A({{12{adc_signed_w[MIXER_WIDTH-1]}}, adc_signed_w}), // Sign-extend 18b to 30b
.B({{2{cos_out[15]}}, cos_out}), // Sign-extend 16b to 18b
.B({{2{cos_nco_pipe[15]}}, cos_nco_pipe}), // Sign-extend 16b to 18b (pipelined)
.C(48'b0),
.D(25'b0),
.CARRYIN(1'b0),
@@ -432,7 +462,7 @@ DSP48E1 #(
.CED(1'b0),
.CEINMODE(1'b0),
.A({{12{adc_signed_w[MIXER_WIDTH-1]}}, adc_signed_w}),
.B({{2{sin_out[15]}}, sin_out}),
.B({{2{sin_nco_pipe[15]}}, sin_nco_pipe}),
.C(48'b0),
.D(25'b0),
.CARRYIN(1'b0),
@@ -492,7 +522,7 @@ always @(posedge clk_400m or negedge reset_n_400m) begin
mixer_overflow_q <= 0;
saturation_count <= 0;
overflow_detected <= 0;
end else if (dsp_valid_pipe[3]) begin
end else if (dsp_valid_pipe[4]) begin
// Force saturation for testing (applied after DSP output, not on input path)
if (force_saturation_sync) begin
mixed_i <= 34'h1FFFFFFFF;