Skip to content

Conducted test (milestone M3)

In a conducted test the signal generator feeds the ESP32 through cables and attenuators. No signal is radiated.

Safety and regulations

  • Never connect an antenna to the generator (HackRF or B210-class USRP). The RF path must be closed: cables and attenuators only.
  • 2483.5–2500 MHz is shared with satellite services such as Globalstar. Transmitting from an antenna in this band needs a radio licence and can interfere with other receivers.
  • Put the receiver in a metal enclosure if possible, to reduce leakage.
  • HackRF: always keep the RF amplifier off (-a 0) and antenna-port power off (-p 0). snappnt.io.generators.hackrf_transfer_cmd always adds both.
  • Never connect a generator directly to the ESP32 without attenuators. It can damage the ESP32 input.
  • snappnt never starts a transmission. It only writes playback files and builds command lines for a person to check and run.
  • Terminate the generator's unused ports with 50 Ω loads (on a B210-class USRP: the other port of the channel in use and both ports of the other channel). Keep a bare board in its case or in a metal box.
  • Connect the whole chain before transmission starts, and stop transmission before disconnecting anything. Start with the full attenuation and the lowest transmit gain.
  • The emission of this bench (leakage from the generator, cables and attenuators) has not been measured against any regulatory limit (issue #57). Whoever transmits, even into a closed cable path, is responsible for complying with the radio regulations where they are.
  • Check the regulations that apply where you are; this page is not legal advice.

Connection

B210-class USRP (TX) ─ 30 dB ─ splitter ─┬─ 30 dB ─ DC block ─ XIAO ESP32C3 (U.FL)
                                          └─ 30 dB ─ reference receiver (USRP)
  • On the XIAO ESP32C3 the U.FL connector goes straight to the chip's LNA input, without an RF switch.
  • The splitter sends the same signal to a reference receiver. If the reference receiver acquires the signal and the ESP32 does not, the problem is on the ESP32 side (capture, gain, or tuning).
  • The attenuator values are a starting point. Begin with 60–70 dB in total.

Without a splitter

When no divider is available (as in the test of Conducted test of NavIC S-band SPS on the XIAO ESP32C3), the generator feeds the ESP32 alone through the attenuators, and a reference receiver records the same playback afterwards in place of the ESP32. This replaces the simultaneous reference only in part: anything that changes between the two recordings is not seen.

Setting the level without a power meter

The noise in the playback must be well above the receiver's own noise (the 10 dB margin of snappnt link-budget, decision D-012). Without a power meter this is checked on the receiver:

  1. Fix the receiver's gain index and analog bandwidth (--gain <index> --bandwidth-mhz <MHz>) for the whole test.
  2. Take 20 captures with the generator off and 20 with the 60 dB-Hz playback on, and compare them with tools/conducted_pd.py rise <off> <on>. It reports the rise of the power spectral density in the generator's band, without the generator's LO leakage line and the bins near 0 Hz, and the fraction of samples at the ends of the converter's range.
  3. Raise the transmit gain until the rise is at least 10 dB and no sample is at the ends of the range.

With a B210 clone, 60 dB of attenuation, gain index 60 and 14 MHz bandwidth on a XIAO ESP32C3, a transmit gain of 60 dB gave a rise of about 18 dB; 0 and 20 dB gave no measurable rise.

Leak check

After the level is set, stop transmission, take the receiver's cable off the last part of the chain (for example the DC block), play the 60 dB-Hz file at the same gain and capture again, and compare with generator-off captures taken in the same arrangement. The in-band rise must be below 1 dB; otherwise the signal reaches the receiver by another path, and the layout must be changed. Also acquire these captures: a leak too weak to raise the power can still be detected by correlation, which matters for the weak-signal test at the end of this page.

Input level and C/N0 calculator

snappnt link-budget adds up the losses between the generator and each receiver, and gives the input level in dBm. It takes no default for the generator power, the losses or the ESP32 noise figure, so each run states them; a path with no loss at all is given as --loss 0, and the command refuses to run when a receiver's path has no loss option. --gen-dbm is the power of the signal alone. When the playback file also contains software-added noise, a power meter at the generator output reads signal plus noise, so that reading must not be passed as --gen-dbm. The splitter's loss on each output is a value you pass: an ideal 2-way split is 3.01 dB, and a real splitter adds some excess loss.

snappnt link-budget --gen-dbm -10 --loss att1=30 --loss splitter=3 --loss att2=30 \
    --loss dc_block=1 --esp32-nf-db 5 --scenario-cn0-dbhz 50

With these example values the input level is −10 − 30 − 3 − 30 − 1 = −74 dBm. A receiver noise figure of 5 dB (an example, not a measured value) gives a noise density of −174 + 5 = −169 dBm/Hz, so a noise-free signal would have C/N0 = −74 + 169 = 95 dB-Hz. This is the weak-signal test of the section "Next step" below.

  • --scenario-cn0-dbhz checks the software-noise test (step 1 of the procedure). The noise in the playback file has density N0_inj = P_in − C/N0 of the scenario. The command prints it next to the receiver's own noise density, and the C/N0 the receiver would actually see with both noises. The check passes when the injected noise is at least --margin-db above the receiver's noise. The default of 10 dB, which gives a C/N0 error of about 0.4 dB, is a choice made for this tool, not a value from a source (decision D-012). Exit code 1 means the check failed.
  • --esp32-loss and --ref-loss list the losses on each branch after the splitter; --ref-nf-db turns on the reference receiver, and --ref-loss without it is an error. --loss is the part of the path shared by both.
  • The noise density uses 290 K (−174 dBm/Hz plus the noise figure), the usual convention for noise figures. Signal power is the power in the received band; quantisation and filter losses inside the receiver are not included.
  • TODO: the noise figures of the ESP32 and of the reference receiver are not verified. A bench measurement would settle them.

Record the inputs and results of each run in the conducted test log template.

Procedure

  1. Write the playback file. snappnt sim scenarios/navic_s_conducted_gen.yaml --uhd --hackrf
    • The file is one second long, which is 50 NavIC data symbols, so looping it does not cut a symbol in half at the loop point.
    • The carrier sits 1.5 MHz above the generator's centre frequency, so the generator's LO leakage (a spike at the centre) does not overlap the signal.
    • Noise is added in software, so the C/N0 is set by the scenario, not by the attenuators. Feed the ESP32 at a comfortable level, around −60 to −70 dBm.
  2. Set the generator's centre frequency. 2492.028 MHz − 1.5 MHz = 2490.528 MHz. The same value is written to core:frequency in the SigMF metadata.
  3. Transmit (a person does this by hand). Build the command with snappnt.io.generators.uhd_tx_cmd(...) or hackrf_transfer_cmd(...), read it, then run it.
  4. Capture with the XIAO. snappnt capture <port> --freq-hz 2492e6 -o out/run1, see "Capturing with snappnt capture" below.
  5. Acquire. snappnt acquire <file> --signal navic_s_sps --prn 10 --freq-span 60000 The crystal errors of the ESP32 and the generator can add up to tens of kHz, so search widely.
  6. Vary C/N0 and compare. Regenerate the playback file with different C/N0 values, measure the detection probability, and plot it over the simulated curve from snappnt sweep scenarios/navic_s_esp32c3.yaml.

Capturing with snappnt capture

snappnt capture <serial port> --freq-hz 2492e6 --rate-sps 80e6 -n 16380 --count 5 -o out/run1
snappnt capture --dry-run --freq-hz 2492e6        # print the commands, send nothing
  • The firmware tunes whole MHz only (100 to 6000). The remainder of the signal's carrier frequency (here 28 kHz for 2492.028 MHz) appears as a frequency offset in processing, so search widely in snappnt acquire.
  • --gain auto (default) selects the hardware AGC (GAIN HARDWARE). --gain <index> sets GAIN MANUAL <index>. The firmware does not report the AGC gain it chose, so snappnt:gain_index is null in AGC mode.

  • Settings are sent once at the start of a run, but another program can change them later: the firmware frees its hold on the radio after 5 s without a command, and a second program on the same serial device (USB or UART) is not refused even while the hold is active (docs/design/espsdr-protocol.md, "Transport"). Close the browser viewer and other serial programs before a run. Whether each recorded setting was confirmed immediately before its capture:

    Setting Metadata keys Confirmed before each capture
    Gain mode and index snappnt:espsdr_gain_reply, snappnt:espsdr_gain_mode, snappnt:espsdr_gain_index Yes, by GAIN?, when the reply is a GAIN line. Otherwise (for example ERR command) the parsed keys are null and the setting is not confirmed. If the reported setting differs from the requested one (snappnt:gain_mode, snappnt:gain_index), a warning naming the capture is printed and the capture is kept.
    Low-pass code snappnt:espsdr_lpf_reply, snappnt:espsdr_lpf_code Yes, by LPF?, when the reply is an LPF line. Otherwise the parsed keys are null and the code is not confirmed.
    Analog bandwidth in MHz snappnt:analog_bandwidth_mhz No. It is the value requested with --bandwidth-mhz (null without it). The firmware reports only the low-pass code, not MHz.
    Frequency core:frequency No. The firmware has no query for the tuned frequency; the value is the one set with FREQ at the start of the run.
    - --bandwidth-mhz sets the analog bandwidth; without it the board's setting is unchanged.
    The firmware keeps the last setting while it is powered, including one made by another
    program such as the browser viewer. snappnt capture therefore sends LPF? immediately
    before every capture and records the reply in that recording. snappnt:analog_bandwidth_mhz holds a value only when --bandwidth-mhz
    was given; otherwise it is null, meaning the bandwidth in MHz is unknown, and the
    low-pass capacitor code in effect is in snappnt:espsdr_lpf_code (see
    docs/design/espsdr-protocol.md, "Analog low-pass setting"). Pass --bandwidth-mhz when
    the bandwidth matters for the recording.
    --bits 8 uses the 8-bit transfer, --bits 10 (default) the packed 10-bit transfer.
    - The command refuses to start if any output file already exists; --overwrite replaces them.
    --dry-run prints the full sequence including the final RELEASE. It checks the frequency,
    the sample rate in the ESP-SDR rate table, the bandwidth (0, or 14 to 62 MHz) and the bit
    depth. The maximum gain index, the rates a particular chip offers and the maximum samples per
    capture are reported by the board (LIMITS?, INFO), so only a live run checks them. A file
    that cannot be written (for example a full disk) ends the run with exit code 1, naming the
    capture. Each recording is written under temporary names and renamed only when both files
    are complete, so a failed write leaves no partial files and an existing recording that
    --overwrite was meant to replace stays intact.
    - With --count N greater than 1 each capture is its own recording, run1_0000, run1_0001,
    and so on (decision D-011). Captures already written stay if a later one fails; a damaged
    capture (wrong CRC-32) ends the run with exit code 1 after the port is resynchronised.
    - Metadata in each .sigmf-meta: core:hw (ESP-SDR and the chip family), snappnt:espsdr_info
    (the firmware's INFO reply), snappnt:host_time_utc and core:datetime,
    snappnt:gain_mode, snappnt:gain_index, snappnt:analog_bandwidth_mhz,
    snappnt:espsdr_lpf_reply, snappnt:espsdr_lpf_code, snappnt:espsdr_lpf_calibrated_codes,
    snappnt:espsdr_gain_reply, snappnt:espsdr_gain_mode, snappnt:espsdr_gain_index,
    snappnt:espsdr_transfer_bits, snappnt:espsdr_capture_us. The serial port name, host name,
    user name and output path are not recorded.
    - Verified on hardware in the conducted test of issue #11
    (Conducted test of NavIC S-band SPS on the XIAO ESP32C3): on one
    XIAO ESP32C3, 2 180 captures kept (of 2 200 taken) with --gain 60 --bandwidth-mhz 14 at 2492 MHz, each with
    GAIN MANUAL 60 0 79 1 and LPF 63 34 34 recorded; the NavIC signal was acquired in the
    captures taken while a signal was played.

Practical notes from the first test

  • Underruns. tx_samples_from_file printed U (underrun) now and then at 8 MSa/s over USB 3. With --args "num_send_frames=512" it printed U only when stopped with Ctrl-C. Record whether U appears during a run; repeat a run with underruns.
  • Warm-up. After the XIAO was plugged in again, its carrier frequency moved by about 5 kHz (two bins of a 0.2 ms capture) within a few minutes. Keep the receiver powered and check that the frequency is stable (for example three sets of 20 captures, two minutes apart, in the same bin) before measuring.
  • Brackets. Take 20 captures at 60 dB-Hz before and after each run; the detection criterion uses their frequency (decision D-022).
  • Confirm transmission. Check that the generator prints its "Press Ctrl + C to stop streaming" line before capturing; a run captured before the generator had started showed no rise and no detection.

Next step: a weak signal without added noise

Feeding a noise-free signal attenuated to around −130 dBm shows the real sensitivity, including the ESP32's own noise figure and any noise folding at low sample rates. At this level leakage from the generator matters, so use several attenuators in series and good shielding.