UE YAML Reference

How to use this page

UE reads UE.yaml from its current working directory at startup. Copy a config/UE_*.yaml preset that matches your hardware and scenario, then edit the copy. Configuration is not hot-reloaded; restart the UE after making changes.

The tables follow the top-level YAML structure and use full paths such as uplink.arq_enabled to distinguish similarly named fields. A typical value is guidance, not a guaranteed default for every preset. If an optional section is omitted, the parser supplies its default values.

Keep coupled frame-structure, duplex, frequency, and resource-mapping fields consistent with the BS configuration.

Parameter tables

radio

Path	Type/Unit	Typical Value	Description
radio.radio_backend	string	uhd	Radio I/O backend. Use uhd for USRPs, sim for the simulator, or xr1710g for the CPU receive-only UE.
radio.xr1710g_shm	string	/xr1710g-live	POSIX shared-memory name of the external single-channel host receiver.
radio.xr1710g_center_freq	float / Hz	2442000000	Required actual source LO, read from the XR1710G controller; not an RF tuning command.
radio.xr1710g_attach_timeout_s	float / s	5	Wait for a live host receiver, within 0–3600 seconds.
radio.xr1710g_pll_command	string	""	Optional persistent PLL helper command, executed on the receiver host. Empty uses software NCO; configured uses hardware LO fine correction within ±100 kHz. See setup below.

See XR1710G setup for supported rates, source ownership and timing limits.

simulation

Path	Type/Unit	Typical Value	Description
simulation.session	string	oisac_sim	Shared simulator session namespace.
simulation.enable_comm_rx	bool	true	Simulator produces the communication RX path.
simulation.enable_sensing_rx	bool	true	Simulator produces sensing RX paths.
simulation.enable_uplink	bool	false	Simulator routes the UE-to-BS uplink stream.
simulation.pacing_enabled	bool	true	Pace simulator output to wall-clock sample time.
simulation.noise_power_dbfs	float / dBFS	-100	AWGN power per RX channel.
simulation.snr_control_enable	bool	false	Maintain target_snr_db by scaling clean simulated signal.
simulation.target_snr_db	float / dB	40	Initial target SNR when SNR control is enabled.
simulation.control_port	int	10002	ChannelSimulator control port.
simulation.cfo_hz	float / Hz	0	Initial BS-to-UE CFO before UE RX correction. The initial UE-to-BS CFO has the opposite sign, is carrier-ratio-scaled in FDD, and its residual then follows UE uplink TX retuning.
simulation.sample_rate_offset_ppm	float / ppm	0	UE sample-clock offset relative to BS.
simulation.timing_offset_samples	int / samples	0	Constant integer sample delay.
simulation.array_spacing_m	float / m	0.04283	Physical ULA spacing.
simulation.array_spacing_lambda	float	0.5	Legacy ULA spacing in wavelengths.
simulation.ring_capacity_samples	int	262144	Shared-memory ring capacity.
simulation.steering_override_file	string	""	Optional array-manifold matrix file. When empty, the simulator generates a ULA manifold from angle_deg. When set, angle_deg no longer affects the array response; encode the angle-dependent amplitude and phase directly in the matrix.
simulation.comm_multipath_taps[]	object[]	optional	Communication taps with delay_samples, gain_db, and phase_deg.
simulation.targets[]	object[]	optional	Monostatic point scatterers.
simulation.bistatic_targets[]	object[]	optional	Bistatic/communication point scatterers.

rf_sampling

Path	Type/Unit	Typical Value	Description
rf_sampling.sample_rate	float / Hz	50000000	Baseband sample rate.
rf_sampling.bandwidth	float / Hz	50000000	Analog bandwidth.
rf_sampling.rx_gain	float / dB	10	UE downlink RX gain.
rf_sampling.rx_agc_enable	bool	true	Enable UE downlink hardware RX AGC.
rf_sampling.rx_agc_low_threshold_db	float / dB	14	Increase RX gain below this filtered delay-spectrum threshold.
rf_sampling.rx_agc_high_threshold_db	float / dB	16	Decrease RX gain above this threshold.
rf_sampling.rx_agc_max_step_db	float / dB	1	Maximum RX gain step per AGC update.
rf_sampling.rx_agc_update_frames	int	4	Minimum processed-frame interval between AGC updates.

usrp_device

Path	Type/Unit	Typical Value	Description
usrp_device.device_args	string	addr=...	USRP device args.
usrp_device.clock_source	string	external	UE clock source: internal, external, or gpsdo.
usrp_device.time_source	string	internal	UE time/PPS source. Empty resolves to internal; pps_phase requires this and clock_source to both be external.

Whenever usrp_device.clock_source is non-empty, CPU, CUDA, and HIP UE perform strict device-time alignment before any stream starts, independently of the selected sync_tracking.ocxo_mode. This includes time_source: internal; time_source selects the PPS input but does not disable startup alignment. Startup observes one PPS transition, arms set_time_next_pps(0), and waits until the device time counter actually resets and time_last_pps=0. A missing PPS, unsupported backend, or failed verification aborts startup. With gpsdo, alignment intentionally replaces any UTC-like device-time epoch with zero. It aligns the software time origin but retains the X310’s 5 ns PPS capture quantization at a 200 MHz master clock.

ofdm_frame

Path	Type/Unit	Typical Value	Description
ofdm_frame.fft_size	int	1024	OFDM FFT size.
ofdm_frame.cp_length	int / samples	128	Cyclic prefix length.
ofdm_frame.sync_pos	int	1	Sync symbol index.
ofdm_frame.enable_sec_sync_symbol	bool	false	Expect a duplicate ZC sync symbol at sync_pos-1.
ofdm_frame.enable_cfo_training_sequence	bool	false	Use the sync_pos+1 CFO training field to resolve CFO aliases.
ofdm_frame.cfo_training_period_samples	int / samples	16	CFO training repetition period.
ofdm_frame.num_symbols	int	100	OFDM symbols per frame.
ofdm_frame.sensing_symbol_num	int	100	Symbols used for sensing.
ofdm_frame.frame_queue_size	int	32	Demodulated RX frame queue capacity.
ofdm_frame.zc_root	int	29	Zadoff-Chu root.
ofdm_frame.pilot_positions	int[]	[571,...]	Comb-pilot subcarrier indices.
ofdm_frame.midframe_pilot_symbols	int[]	[]	Optional mid-frame BPSK pilot symbols.
ofdm_frame.midframe_pilot_seed	int	1296453708	Deterministic mid-frame BPSK pilot seed.

cuda

Path	Type/Unit	Typical Value	Description
cuda.cuda_precision_profile	string	fp32	fp32 preserves the existing path. mixed_fp16 uses FP16 for bulk OFDM FFT/IFFT and GPU LDPC messages while retaining FP32 for channel estimation, tracking, equalization, sensing, and LLR generation. Unsupported FP16 FFT shapes fall back to FP32.
cuda.cuda_demod_pipeline_slots	int	3	CUDA demodulation pipeline slots.
cuda.cuda_ldpc_decoder_backend	string	gpu	CUDA demod LDPC decoder backend: gpu or cpu.
cuda.cuda_ldpc_worker_buffers	int	3	CUDA LDPC async worker batch buffers.
cuda.cuda_ldpc_cross_frame_flush_frames	int	2	Max frames accumulated before CUDA LDPC batch decode.
cuda.cuda_ldpc_cross_frame_flush_us	float / us	1000	Max CUDA LDPC cross-frame batch wait time.

ldpc

Path	Type/Unit	Typical Value	Description
ldpc.decoder_iterations	int	12	Maximum CPU LDPC iterations, at least 1; syndrome early stopping stays enabled.
ldpc.nms_normalization_factor	float	0.75	CPU NMS normalization factor, finite and in (0, 1]; Q16 requires multiples of 0.125.
ldpc.fixed_point	bool	false	Use int16/Q16 CPU LDPC decode path.
ldpc.fixed_point_scale	int	16	LLR scale before int16 saturation.

These fields configure CPU decoding on the UE downlink, for both float32 and Q16. Omitting them selects 12 iterations and a factor of 0.75. Explicit YAML values remain in effect; update both fields to use these defaults in an existing config. Edit them under LDPC decode in the Web Config Console, save, and restart the backend. Q16 supports 0.125, 0.25, 0.375, 0.5, 0.625, 0.75, 0.875, 1.0; the LLR scale applies only when fixed_point: true. More iterations may improve error correction while increasing computation for weak signals.

downlink

Path	Type/Unit	Typical Value	Description
downlink.center_freq	float / Hz	2400000000	UE downlink RF center frequency.
downlink.rx_wire_format	string	sc16	UE downlink RX wire format.
downlink.rx_channel	int	0	UE downlink RX channel index.
downlink.equalizer_mode	string	mmse	Downlink equalizer inverse mode: zf or mmse.
downlink.channel_tracking_mode	string	pilot_phase	Per-symbol comb-pilot tracking mode.
downlink.equalizer_mag_floor	float	1e-6	Lower bound for `
downlink.channel_tracking_min_pilot_snr	float	1e-4	Minimum comb-pilot residual weight before fallback.
downlink.arq_enabled	bool	false	Enable downlink ARQ on the UE receiver.
downlink.arq_ordered_delivery	bool	false	Buffer downlink packets for in-order UDP delivery.
downlink.arq_window_packets	int	256	Downlink ARQ receive/reorder window; valid range 1–32767.
downlink.arq_feedback_interval_ms	int / ms	10	Minimum interval between downlink ARQ ACK feedback packets.

uplink

Path	Type/Unit	Typical Value	Description
uplink.enabled	bool	false	Master switch for UE uplink/duplex.
uplink.duplex_mode	string	tdd	Must match BS: tdd windowed uplink or fdd full-frame uplink.
uplink.center_freq	float / Hz	2500000000	FDD-only uplink carrier. TDD uses the downlink center frequency.
uplink.idle_waveform	string	random_qpsk	UE uplink idle waveform: random_qpsk or zero.
uplink.symbol_start	int	90	TDD-only first uplink symbol.
uplink.symbol_count	int	10	TDD-only uplink window length.
uplink.guard_symbols	int	1	TDD-only leading guard symbols.
uplink.tx_gain	float / dB	0	UE uplink TX gain.
uplink.tx_channel	int	0	UE uplink TX channel index.
uplink.wire_format_tx	string	sc16	UE uplink TX wire format.
uplink.ue_timing_advance	int / samples	50	UE uplink transmit timing advance.
uplink.debug_self_channel	bool	false	Estimate UE self-TX leakage channel from RX windows for TADV debugging.
uplink.ertm_to_enable	bool	false	Enable eRTM TO payload consumption and TO logs.
uplink.ertm_timing_metric	string	delay_magnitude	Differential-TO metric: phase-robust delay_magnitude or coherent maximum_likelihood.
uplink.ertm_delay_oversample_factor	int	10	eRTM delay-spectrum IFFT oversampling factor.
uplink.ertm_dl_rf_delay_samples	float / samples	67.0	Calibrated downlink RF-chain delay for eRTM equations.
uplink.ertm_ul_rf_delay_samples	float / samples	67.0	Calibrated uplink RF-chain delay for eRTM equations.
uplink.ertm_debug_output_enabled	bool	false	Enable UE-side eRTM debug ZMQ spectra.
uplink.ertm_report_interval_frames	int / frames	32	BS eRTM report cadence; keep matched with BS.
uplink.arq_enabled	bool	false	Enable uplink ARQ on the UE transmitter.
uplink.arq_window_packets	int	256	Uplink ARQ outstanding packet window; valid range 1–32767.
uplink.arq_retransmit_timeout_ms	int / ms	100	Uplink ARQ retransmission timeout.
uplink.arq_max_retries	int	5	Max uplink retransmission retries; 0 means unlimited within the window.

With timed uplink enabled on a real radio, CPU, CUDA, and HIP UE startup queries the device’s actual RX and TX sample rates. A difference greater than 1 Hz aborts startup before worker threads are launched because exact uplink identity mapping requires both directions to use the same sample-time grid.

sync_tracking

Path	Type/Unit	Typical Value	Description
sync_tracking.sync_queue_size	int	32	Sync-search batch queue capacity.
sync_tracking.sync_cfo_alias_search_range_hz	float / Hz	800000	CFO range covered by the sync alias resolver.
sync_tracking.reset_hold_s	float / s	0.5	Consecutive invalid-delay or preamble-mismatch duration before a hard reset to sync search.
sync_tracking.software_sync	bool	true	Enable software sync tracking.
sync_tracking.predictive_delay	bool	true	Use CFO-based predictive delay compensation.
sync_tracking.hardware_sync	bool	false	Enable hardware OCXO discipline.
sync_tracking.hardware_sync_tty	string	/dev/ttyUSB0	Serial device for the OCXO DAC controller.
sync_tracking.ocxo_mode	string	auto	auto, freq, and phase retain the legacy TO-based behavior; pps_phase explicitly selects PPS phase locking.
sync_tracking.ocxo_dac_ppm_per_lsb	float	9.71e-7	Calibrated DAC ppm per LSB (MV197 map).
sync_tracking.ocxo_dac_zero_word	int	493367	DAC word for 0 ppm.
sync_tracking.ocxo_dac_state_file	string	.openisac_ocxo_dac_state.yaml	Persist the last SF word acknowledged by the DAC. A relative path is resolved from the UE working directory; an empty value disables restart persistence.
sync_tracking.ocxo_max_abs_ppm	float / ppm	0.45	Absolute OCXO command clamp.
sync_tracking.ocxo_actuator_max_slew_ppm_per_s	float / ppm/s	0.01	Hard actuator slew limit; applied steps are fed back to the controller.
sync_tracking.ocxo_coarse_*	various	see presets	One-target startup correction: Mode A averages residual CFO; Mode B regresses Alignment-continuous TO_clock against absolute frame sequence over the coarse window.
sync_tracking.ocxo_fll_fast_*, ocxo_fll_slow_*	various	see presets	Fast/slow FLL on residual CFO in Mode A or rolling TO_clock regression SFO in Mode B, with threshold hysteresis and dwell.
sync_tracking.ocxo_fll_slow_kp	float	0.1	Slow FLL proportional gain.
sync_tracking.ocxo_fll_slow_ki	float	0.007	Slow FLL integral gain.
sync_tracking.ocxo_fll_slow_max_slew_ppm_per_s	float / ppm/s	0.005	Slow FLL command slew limit.
sync_tracking.ocxo_fll_slow_enter_abs_ppm	float / ppm	0.0005	Residual threshold for entering Slow FLL from Fast FLL.
sync_tracking.ocxo_phase_setpoint_mode	string	freeze_on_lock	Legacy auto/phase target: freeze TO_clock on Phase entry, use zero, or use a fixed target.
sync_tracking.ocxo_phase_setpoint_fixed_samples	float / samples	0.0	Legacy fixed TO_clock target when the setpoint mode is fixed.
sync_tracking.ocxo_phase_*	various	see presets	Velocity-form Phase PI. Its error is setpoint-TO_clock in legacy auto/phase, or wrapped P_U-P_B in pps_phase; entry/exit still uses TO_clock SFO regression.
sync_tracking.ocxo_pps_initial_ap_enable	bool	true	Legacy key name. In pps_phase, start one PPSOUT alignment transaction on every FllSlow -> Phase entry before releasing SF phase tracking. Disabling it emits a startup warning because a large residual may converge very slowly through the slew-limited PI.
sync_tracking.ocxo_pps_initial_ap_max_abs_ns	float / ns	500000000	Absolute AP clamp, matching the PPSDO hardware range.
sync_tracking.ocxo_pps_initial_ap_settle_s	float / s	2.0	Minimum AP settling time; a newer eRTM/PPS observation is also required.
sync_tracking.ocxo_pps_initial_ap_command_timeout_s	float / s	3.0	Total AP worker wall-clock budget, including all JAMSYNCING retries. The aligner waits a 0.5 s terminal-status grace; no AP copy is sent after the worker deadline.
sync_tracking.ocxo_pps_initial_ap_max_retries	int	1	Bounded retries (0..3) when AP was accepted but no fresh post-AP publication arrived by settle_s + command_timeout_s. A retry waits for a new valid publication and recomputes AP from its residual.
sync_tracking.ocxo_pps_initial_ap_deadzone_ns	float / ns	10.0	Skip AP and the settling wait when the PPS error at Phase entry is already within this range.
sync_tracking.ocxo_pps_initial_ap_polarity	int	1	AP command sign for a positive P_U-P_B error. The default is the polarity confirmed by the PPSOUT bench measurement; -1 remains available for different hardware/firmware polarity.
sync_tracking.cfo_ewma_enable	bool	true	EWMA pre-filter on residual CFO ppm for Mode A Coarse/FLL.
sync_tracking.cfo_ewma_alpha	float	0.02	EWMA weight on new residual samples (0..1; higher = faster, noisier).
sync_tracking.desired_peak_pos	int	20	Target delay-peak position for alignment logic.

The hardware presets use ocxo_fll_slow_kp=0.1, ocxo_fll_slow_ki=0.007, ocxo_fll_slow_max_slew_ppm_per_s=0.005, and ocxo_fll_slow_enter_abs_ppm=0.0005. They retain ocxo_fll_slow_deadzone_ppm=5.0e-5, ocxo_phase_kp=2.0e-4, ocxo_phase_ki=1.0e-6, and ocxo_phase_enter_sfo_abs_ppm=2.0e-4. The PPSOUT AP correction is enabled on every Phase entry and uses the bench-confirmed positive command polarity by default.

Normal synchronization reacquisition resets OCXO observations, dwell timers, and loop stages without writing the DAC zero word. The discipliner restarts at the actuator’s retained command. The actuator rebases its slew timer at the first actual nonzero post-lock correction, so time spent searching cannot inflate that correction. After an SF command receives OK, the non-real-time serial worker durably updates ocxo_dac_state_file. On the next UE/CUDAUE process start, that acknowledged raw word initializes both the DAC and controller instead of starting at zero. During graceful shutdown, the serial worker is stopped first and the last SF word that actually received OK is saved again; a newer unacknowledged queued word is never promoted to the steady-state restart value. Restore also requires the saved serial device, zero_word, and ppm_per_lsb to match the active configuration. A missing file starts from the calibrated zero word and is created after an acknowledged command. A corrupt, mismatched, or out-of-range file also falls back to zero, but remains unchanged and disables persistence for that process run; fix or remove it before restarting. This assumes the external DAC itself remains powered and retains its word while the UE process restarts.

auto and phase preserve the existing Phase anchor: the PI controls setpoint - TO_clock, including freeze_on_lock, zero, and fixed. They do not read time_last_pps() or run the AP gate.

pps_phase is a separate explicit mode. For downlink frame N, eRTM reconstructs r_B=B_N-P_B from the BS anchor refreshed immediately before each ERTM5 report is packed, and obtains r_U=U_N-P_U from the UE frame-boundary and PPS timestamps. Its Phase error is wrap_1s(tau_DL_RF - TO_UE + r_B - r_U) = wrap_1s(P_U-P_B), with target zero. This mode is rejected unless sync_tracking.hardware_sync=true, uplink.ertm_to_enable=true, and both usrp_device.clock_source and usrp_device.time_source are explicitly external; an empty time_source means internal. The BS must likewise use the PPSDO-connected external clock and time sources when producing the PPS anchor. The BS anchor refresh runs at the eRTM report cadence configured by uplink.ertm_report_interval_frames, outside the TX/modulation real-time path. On every FllSlow -> Phase transition, the AP gate is rearmed and computes one correction from the latest valid PPS residual. It remains done throughout that Phase residence, so normal Phase updates and Freeze do not repeat AP. If Phase falls back through FLL and later re-enters, one new correction is allowed. AP changes PPSDO PPSOUT only and never resets USRP time, reanchors the air-frame clock, clears eRTM state, or rebuilds TO. If the previous AP is still queued or in flight, the new entry waits for hardware Idle instead of treating it as a terminal queue failure. JAMSYNCING retries reuse the same AP request.

sensing

Path	Type/Unit	Typical Value	Description
sensing.sensing_delay_correction_mode	string	los_tracking	Bistatic sensing delay correction source: los_tracking or ertm_absolute.
sensing.bi_enabled	bool	true	Enable bistatic sensing processing.
sensing.bistatic_phase_reference	bool	false	LoS phase reference; improves sensing with less stable clocks. Requires LoS and a transmitter/receiver pair stationary relative to each other; restart UE to apply.
sensing.phase_reference_min_coherence	float	0.05	Minimum reference coherence in (0, 1]; effective when phase reference is enabled.
sensing.range_fft_size	int	1024	Range FFT size.
sensing.doppler_fft_size	int	100	Doppler FFT size.
sensing.view_range_bins	int	0	Backend RD view width; 0 means full range.
sensing.view_doppler_bins	int	0	Backend RD view height; 0 means full Doppler size.
sensing.output_mode	string	dense	dense full-buffer output or compact_mask selected-RE output.
sensing.on_wire_format	string	complex_float32	Sensing payload wire format.
sensing.backend_processing_enabled	bool	false	Publish backend RD/CFAR/micro-Doppler sidecars when supported.
sensing.symbol_stride	int	20	Default dense-mode STRD applied at startup.

LoS phase reference

This option can improve bistatic sensing coherence and Doppler stability when independent or less stable transmitter/receiver clocks leave residual common phase drift after synchronization. It requires a stable line-of-sight (LoS) path between the transmitter and receiver, and the two radios must remain stationary relative to each other. Sensing targets may still move. Do not enable it for NLoS-only links or when the transmitter and receiver move relative to each other: motion of the reference path would also be subtracted from the measured Doppler.

On the UE Linux host, merge these fields into the active UE.yaml:

sensing:
  bi_enabled: true
  output_mode: dense
  sensing_delay_correction_mode: los_tracking
  bistatic_phase_reference: true
  phase_reference_min_coherence: 0.05

This is a startup YAML option, also editable in the Web Config Console. Save the YAML and restart the UE to apply it; there is currently no online enable/disable command for this option.

The correction removes residual common phase from bistatic CSI in dense output mode after CFO/SFO/delay and system-response correction, before MTI.

Each slow-time row uses a Hamming-weighted LoS reference and a unit-magnitude conjugate rotation on all subcarriers. The reference follows the known ALGN display translation: it cancels that phase ramp only while measuring the reference, and retains the requested delay shift in the output. A translation clears the incomplete sensing batch and MTI/micro-Doppler history so the two display coordinates are not mixed. Amplitudes are preserved. The direct path must be static and dominant: the displayed Doppler is relative to this path; a moving reference also subtracts its own motion. This does not correct the radio sample clock.

sensing.phase_reference_min_coherence defaults to 0.05, with range (0, 1]. Coherence is |sum(w H)|² / (sum(w) sum(w |H|²)), where w is the Hamming weight and H is evaluated after cancelling the known display phase ramp. A weak, near-zero or non-finite reference rejects the entire sensing batch and clears MTI/micro-Doppler history. Communication samples are unaffected. Missing slow-time samples reset the sensing window and history to keep Doppler processing on a uniform timeline.

Restart the UE after changing these YAML fields. On the UE Linux host, check the startup log for static LoS phase reference enabled before MTI. Frequent weak/invalid LoS phase reference messages indicate that this path is unsuitable as a reference; disable the feature or improve the direct-path signal.

resource_preview

Path	Type/Unit	Typical Value	Description
resource_preview.data_resource_blocks[]	object[]	optional	Payload / sensing-pilot RE rectangles; each item has kind, symbol_start, symbol_count, subcarrier_start, and subcarrier_count.
resource_preview.mask_blocks[]	object[]	optional	Compact sensing RE rectangles with symbol_start, symbol_count, subcarrier_start, and subcarrier_count.

measurement

Path	Type/Unit	Typical Value	Description
measurement.measurement_enable	bool	false	Enable internal PRBS measurement checking.
measurement.measurement_mode	string	internal_prbs	Measurement checker mode.
measurement.measurement_run_id	string	""	Run ID written to measurement CSV summaries.
measurement.measurement_output_dir	string	""	Output directory for measurement CSV summaries.
measurement.measurement_payload_bytes	int / bytes	1024	Expected bytes per measurement payload.
measurement.measurement_prbs_seed	int	0x5A	Base seed for rebuilding deterministic PRBS payloads.
measurement.measurement_packets_per_point	int	1	Expected packets for one measurement epoch.
measurement.measurement_max_packets_per_frame	int	1	Max measurement packets checked per frame; 0 means unlimited.

network_output

Path	Type/Unit	Typical Value	Description
network_output.udp_input_ip	string / IPv4	0.0.0.0	UE uplink payload UDP bind IP.
network_output.udp_input_port	int	50002	UE uplink payload UDP bind port.
network_output.udp_output_ip	string / IPv4	""	UE decoded downlink UDP destination IP; empty uses runtime.default_out_ip.
network_output.udp_output_port	int	50001	UE decoded downlink UDP destination port.
network_output.udp_egress_pacer_enabled	bool	false	Enable queued pacing for decoded UDP egress.
network_output.udp_egress_pacer_target_mbps	float / Mbps	0	Egress pacer target rate; 0 auto-estimates.
network_output.udp_egress_pacer_queue_packets	int	10240	Egress pacer queue capacity.
network_output.udp_egress_pacer_max_delay_ms	float / ms	0	Max queued packet age; 0 disables age drops.
network_output.bi_sensing_output_enabled	bool	true	Enable bistatic sensing ZMQ output.
network_output.bi_sensing_ip	string / IPv4	0.0.0.0	Bistatic sensing/control ZMQ bind IP.
network_output.bi_sensing_port	int	8889	Bistatic sensing PUB port.
network_output.channel_ip	string / IPv4	0.0.0.0	Channel-estimate PUB IP.
network_output.channel_port	int	12348	Channel-estimate PUB port.
network_output.pdf_ip	string / IPv4	0.0.0.0	Delay-spectrum PUB IP.
network_output.pdf_port	int	12349	Delay-spectrum PUB port.
network_output.constellation_ip	string / IPv4	0.0.0.0	Constellation PUB IP.
network_output.constellation_port	int	12346	Constellation PUB port.
network_output.self_channel_ip	string / IPv4	0.0.0.0	UE self-channel debug PUB IP.
network_output.self_channel_port	int	12350	UE self-channel debug PUB port.
network_output.self_pdf_ip	string / IPv4	0.0.0.0	UE self-delay-spectrum debug PUB IP.
network_output.self_pdf_port	int	12351	UE self-delay-spectrum debug PUB port.
network_output.ertm_debug_ip	string / IPv4	0.0.0.0	UE eRTM debug PUB IP.
network_output.ertm_debug_port	int	12362	UE eRTM debug PUB port.
network_output.ocxo_debug_ip	string / IPv4	0.0.0.0	OCXO debug ZMQ PUB bind IP.
network_output.ocxo_debug_port	int	12347	OCXO debug ZMQ PUB port.
network_output.control_port	int	10001	ZMQ ROUTER port for runtime control.

cpu_cores

Path	Type/Unit	Typical Value	Description
cpu_cores.downlink_cpu_cores	int[]	[1,2,3]	UE downlink cores: RX, processing, bit processing.
cpu_cores.demod_worker_cpu_cores	int[]	[]	UE CPU demod worker cores; empty starts one unbound worker.
cpu_cores.ldpc_worker_cpu_cores	int[]	[]	UE CPU LDPC decode worker cores; empty starts one unbound worker.
cpu_cores.sensing_cpu_cores	int[]	[4]	UE bistatic sensing cores.
cpu_cores.uplink_cpu_cores	int[]	[]	UE uplink cores: LDPC encode, modulation, TX send, UDP receive.
cpu_cores.main_cpu_core	int	-1	Main-thread CPU core; -1 disables binding.

runtime

Path	Type/Unit	Typical Value	Description
runtime.default_out_ip	string / IPv4	127.0.0.1	Default destination IP for UDP outputs when a specific IP field is empty.

When hardware sync is enabled, the OCXODBG3 ZeroMQ packet retains these 16 fixed channels:

Channel	Signal	Channel	Signal
0	modulo CFO (Hz)	8	active-window TO_clock regression SFO (ppm)
1	de-aliased CFO (Hz)	9	cumulative Alignment (samples)
2	raw residual CFO (ppm)	10	Alignment feedforward (samples)
3	filtered control CFO (ppm)	11	desired OCXO delta (ppm)
4	OFDM delay-track SFO (ppm; channel compensation only)	12	applied OCXO delta (ppm)
5	raw TO_UE (samples)	13	applied OCXO command (ppm)
6	continuous TO_clock (samples)	14	coarse target outstanding (ppm)
7	active Phase error: legacy setpoint-TO_clock or pps_phase P_U-P_B (samples)	15	stage (0..4: Coarse, Fast, Slow, Phase, Freeze)

The packet is a fixed 152-byte little-endian frame: magic, version, channel count, sequence, Unix-nanosecond source timestamp, the original 16 float32 values, and a typed PPS diagnostic trailer. The trailer and CSV append r_bs_samples and r_ue_samples (float64), bs_last_pps_ticks, bs_t0_to_pps_ticks, and ue_last_pps_ns (exact int64), plus bs_tick_rate and sample_rate (float64). These fields distinguish BS-anchor, UE-PPS-latch, and TO discontinuities without passing absolute timestamps through float32. The viewer still accepts legacy OCXODBG2 packets. Launch the rolling viewer with python3 scripts/plot_ocxo_debug.py. Use --channels 6,7,8,13,15 to choose initial curves and --record capture/ocxo.csv to start CSV recording; automatic names use ocxo_debug_YYYYMMDD_01.csv and increment to _02, _03, and so on for later captures on the same day. Recording always stores all channels even when only a subset is visible. Periodic state packets are supplemented by immediate control, stage-transition, TO-jump, and Alignment event packets so one-frame delta channels are not lost.

logging

Same hierarchical filter as BS (logging.*). Debug examples: ertm: debug, recovery: info, demod.snr: info. Performance (paths end with _profiling, off until set): demod_profiling: info, demod_eq_profiling: info.

Resource-map notes:

• resource_preview.data_resource_blocks should normally match between BS and UE, including kind.
• Built-in ZC sync symbols, the optional CFO training field, comb pilots, and mid-frame BPSK pilots take precedence over configured resource blocks.
• resource_preview.mask_blocks controls compact sensing export only. In output_mode=compact_mask, runtime STRD is ignored because the mask defines the sampling pattern.
• Compact sensing payloads begin with CompactSensingFrameHeader, followed by fixed-order raw complex<float> samples.