Capturing long stream of pulses with RMT?

[OSCPUDEV]

I have an application where I need to capture just the time between low-going pulses on a GPIO. The pulses come from a sensor that is proportional to the event it is monitoring. The pulse width can range from 2.000us to 20.000us, and has a 25ns resolution.

It seems that the RMT peripheral should be able to capture up to a 12.5ns resolution with CLK_DIV=1 (2 would be 25ns), but the number of samples is limited to some buffer that is built into the periphera (64 x 32 for each block)l. I need to capture 300,000 samples at a time (and will need to play those back, but I will leave that for later). There is 2MB of PSRAM in the ESP32S2 that I am using, which is plenty of space to hold the samples.

Is it wrong to believe that it should be possible to start a capture and then copy the 32 bit capture times into the PSRAM? I don't see a hardware based ping-pong where it would generate an interrupt after every buffer switch. I only see an interrupt for every single pulse capture. I know that the PSRAM is theoretically capable of storing the data fast enough, but is 2us too short of a period for handling the interrupt every single capture (even if using assembly code)?

Is there a better solution for this than the RMT?

I would appreciate any feedback on this. Thanks!

[ESP_Sprite]

From memory, the ESP32 cannot do that, but later chips (at least the S3 and C3/C6) have the capability to do ping-pong and the S3 has the capability to DMA directly from RMT to memory.

[OSCPUDEV]

Ok, is it possible to either use the RMT driver or install your own ISR that can move data fast enough to local memory and then to PSRAM? I also need to have some way to know when the buffer is full or maybe move each byte in the ISR?

I might be able to use a ESP32S3 if there are enough GPIOs. I just need a solution that will be able to capture the sensor stream fast enough, and capture a few seconds worth of data.

[OSCPUDEV]

Is there any more details (and examples) on the RMT for the ESP32-S3?

It seems that you can choose between storing to a 'block of memory' and DMA'ing to memory.

Is the driver just doing it's own ping-pong and storing to any memory? Can that memory be the PSRAM memory?

It seems that the latest RMT driver for the ESP32-S3 breaks virtually everything in the previous driver structures. Is the new driver more efficient? I know that the RMT transmit (for any ESP32 version) can easily handle sub-microsecond output pulses (RGB LEDs are a good example of this). Can the receive handle pulses as short as 1.000us? My input signal has a 1.000us high-low-high pulse, with the next pulse occurring sometime later that represents the actual sensor time (ie. a 4.250us sensor time would be a 1.000us high-low-high pulse followed by 3.250us duration of high until the next pulse).

Maybe RMT is not the way to do this? I just need to measure the time between every low going pulse and store those times, really as a 16 bit value since I will never see anything larger than 20.000us.

[ESP_Sprite]

Examples are available within ESP-IDF. The new driver is indeed quite different from the old one, but this needed to happen in order to support things like DMA; I don't know if the old driver supports that. Resolution of RMT Rx is the same as RMT Tx; with an 80MHz APB bus, you should be able to capture pulses down to 12 ns if you configure the hardware to do so.

[OSCPUDEV]

Yeah, it seems that the hardware has a max resolution of 12.5ns which is double of what I actually need. I don't really need DMA, and I doubt that will work for storing directly into PSRAM anyways. I just need a simple way to measure the time between a lot (300K) fast low-going pulses and store those times. PSRAM is needed because of the shear number of samples that need to be stored.