SOC_RMT_MEM_WORDS_PER_CHANNEL, which is used in PulseIn.c, varies per processor. However, on the ESP32, it is 64, and on ESP32-S3 and C6, it is 48. Not sure what to make of that, given your test results above.

Its been a little while since I wrote this code. I think my intent was to limit the maxlen to the amount the RMT peripheral can buffer. I didn't want to assume PSRAM was accessible when PulseIn was running. (Writing the filesystem will disable it temporarily.)

I think some RMTs can use multiple channels worth of memory if needed. That could improve things.

Note each RMT word stores two "pulses" one active and one inactive. That's why 64 words give a length of 128.

@dhalbert

Thanks for your comment.

This is a difficult area to understand for me, but I found an article that might be relevant. It seems that chips with the feature SOC_RMT_SUPPORT_RX_PINGPONG can handle more data than buffers.

https://esp32.com/viewtopic.php?t=24484

Not to muddle up this topic but I also noticed a maxlen limit of 128 pulses on an ESP32-S3 QTPy no PSRAM. My main issue was that PulseIn doesn't work at all to capture a set of pulses (<128) from a constant pulsed data stream. Details here

So how to get continuously pulses and store them in an array without missing any?

Raise an exception to note the limitation.