Hi Matteo,
we are building something close to your sounder — a hand-held array of ten ESP32-C5 receivers whose CSI is combined — and we hit the same wall with a shared 48 MHz clock. We have no boards on the bench yet, so all of this is reading and physics, not measurement; but it may save you some time, and the second half may spare you the clock altogether.
David is right that Espressif’s checklist calls the crystal compulsory and the firmware supports only 48 MHz. We read that as the frequency, though, not a ban: a clean 48 MHz on the oscillator’s input should look to the firmware like its crystal, and Espressif’s own esp-crab runs two C5s on one crystal. What is unknown is the hardware, so that is what the points below are about.
If you keep the shared clock
One crystal will not drive eight chips. Hung on the master’s crystal node, seven more pin capacitances, seven 24 nH inductors and the traces between them pull its swing below what the chips need (Espressif asks for a crystal amplitude above 500 mV), or stop it. Espressif’s own esp-crab (esp-csi, examples/esp-crab) shares one crystal between two C5s only. For eight: a 48 MHz TCXO into an LVCMOS fan-out buffer, one output per chip, series-terminated, AC-coupled, short traces — and the clock running before CHIP_PU rises.
Drive the amplifier’s input. XTAL_P/XTAL_N are a Pierce oscillator; the datasheet (v1.5, Table 2-9) does not say which pin is the inverter’s input — only “External clock input/output connected to chip’s crystal or oscillator. P/N means differential clock positive/negative”. With a scope: on a working board the input usually carries the smaller, cleaner sine (use the lowest-capacitance probe you have); with the crystal removed, try ~100 pF into XTAL_P, then XTAL_N, then both in antiphase, and see which boots.
Keep the 24 nH on XTAL_P. Espressif puts it there to keep the crystal’s harmonics off the radio; at 48 MHz it is ~7 ohms and does not stop the clock. An injected square clock needs it more — its 51st harmonic is 2448 MHz.
A shared clock gives a common frequency, not a fixed phase. The ESPARGOS authors (arXiv 2502.09405) report that the phase changes after every reset and every change of channel. So a calibration after each lock is needed whatever the clock: a common reference signal (ESPARGOS), one element sending to the rest, or — if your array moves — fitting each receiver’s fixed phase out of the data. Power save off and no retune during a measurement keep one calibration good.
If amplitude is enough: no shared clock
That is where we ended. With every chip on its own crystal the phase between receivers is new in every frame, but the amplitudes, with the gain held fixed, carry nearly everything we need:
on each frame, set the receivers against one another — the sender’s power and the room’s fading cancel;
let the sender alternate two crossed antennas a frame each: in one receiver the ratio of the two frames has no receiver gain in it at all;
if the array moves, level each receiver against its own record over the motion;
for focusing, use the sender’s own wave as the reference of an in-line (Gabor) hologram: the change in amplitude is the part of the scattered wave in step with the illumination, and it can be propagated back.
Our model of this (rough: a 2D wall, single frequencies, assumed materials) draws a cable a few centimetres deep in a wall about as well as with full phase; what it loses is depth. It is open, with the scripts — principle.md and calc/nosync.py in the repository below. For angle-of-arrival or true MIMO you will need the phase, so this may not fit you, but if part of your work is sensing it may spare you the clock.
What else is published about clocks on these chips: ESPARGOS’s docs (“Phase-Coherent Multi-Array Systems”) — a 40 MHz reference on a coax shared with a 2.4 GHz phase signal, a star of identical cables from a resistive splitter, an amplifier beyond two boards, “1 Vpp at the SYNC IN … is definitely sufficient”, and a slave without the clock “cannot boot its sensors” (their chips are 40 MHz ones, not the C5). The esp-crab hardware files are not out; GitHub issue #259 on esp-csi asks for them.
If you try the checks in point 2, we would be glad to hear what you find — it would settle the question for everyone building such an array on the C5. And if we get there first, we will post it here.
Best regards,
Pavel
Project-NIC — GitHub - Project-NIC/NIC-Sluchalo: A hand-held wall scanner: what is in a wall, seen by the shadow Wi-Fi casts through it, with a knock and microphones for masonry. A concept — nothing built yet. · GitHub