0.5 MSPS external ADC

[Stabilo]

Hello,

I want to use a STM32F7 to retrieve ADC measurements from an external ADC over SPI (16-bit read operation to retrieve the measurement).

In theory, it should be possible.

My question is: how can I retrieve the measurements minimizing the host processor load ?

My idea is to ;

• use a 48 Mhz SPI
• use an interrupt to trigger SPI read.
• use SPI read operation with DMA to store the data in buffer.

The problem is that my interrupt will fire each 2us ... I don't think this is reasonable on a microcontroller.

Is there a smarter way to do that on a MCU ?

Or should I choose another solution like FPGA or DSP ?

Thanks !

[TDK]

Adjust your SPI clock such that it reads 16 bits every 2us and have it continuously read, then use DMA to transfer many values at once. You can use half- and full-complete callbacks to process the data.

[Stabilo]

Thank you TDK.

This is a great idea, but my ADC is a bit particular.. I need to trigger the ADC measurement by a GPIO rising edge. I need to see if I can continuously "poll" the SPI bus,

[TDK]

It's still possible, a little complicated.

Use a master timer to trigger the ADC.

Use a slave timer in gated mode (off of master timer) to generate the 16 SCK clocks.

Use SPI slave in receive mode to get data.

[Stabilo]

Thank you !

If I understand correctly, the slave timer allows to generate a kind of "delayed" interrupt, and is this interrupt I can read data through SPI ?

[TDK]

I wouldn't use interrupts on each sample. That's going to be too many. Set up the SPI to transfer into a large buffer using DMA, possibly a circular buffer, then process it at some later time, perhaps during the half- and full-complete interrupts, which will be generated by the DMA reaching a certain number of transferring values.
The slave timer is just used to generate the SCK clock. Set it up as a 50% PWM, set up the master timer to be high for exactly 16 clock cycles.

[Stabilo]

The only point that I don't understand is: the SPI peripheral is already generating a clock signal right ? Maybe there will be a conflict between the SPI generated clock and the one generated by the slave timer ...