Always wanted to have Msps ADCs on a Raspberry ... Let's dig into it.
Already have an account? Log in.
To make the experience fit your profile, pick a username and tell us what interests you.
The idea here lies on the idea that we haven't tapped yet into the full potential of the raspberry pi in terms of ADC.
Some have shown that the first gen of Raspberry could go to 10Msps as detailed here https://digibird1.wordpress.com/raspberry-pi-as-an-oscilloscope-10-msps/
My take is that the new Raspberry Pi.es (3 or even W) can surely go above. And I want to try it, either with this old CA3306E or with more recent kick-ass ADCs.
Let's see!
I've finally got a first batch from the fab, and got a couple of them on tindie for experiments! https://www.tindie.com/products/kelu124/20msps-adc-cape-for-raspberry-pi/
Yeah, that worked! Ultimately, I'm getting 20k points in 840us, so an equivalent of 24Msps. See below for more analysis on a signal at around 1.9 MHz:
See the full log at https://github.com/kelu124/echomods/blob/master/elmo/data/20170713-TwoADCs.ipynb -- thanks to macrofab for their perfect PCBA role =)
TLDR:
I was still able to get a good ultrasound image with this ADC (see my other project on ultrasound imaging dev kit blog post):
For general info, here's what the GPIO memory looks like during 10k samples at 10MHz =)
and here's what 2 ADCs normally interleaved produce when one the ADCs has two output pins soldered together:
At least I still have one ADC to play with, and experiment!
Problem and solution!
Problem was that the Pi0 or PiW, with the current kernel module, can't run faster than around 10 to 12 Msps. Clock can be faster, but the memcpy limits the transfer to memory. Soooo.. what can we do? As a reminder, I wanted at least a 10Msps 10bit ADC.
Solution Why not interleaving two ADCs, with clocks in opposite phase? Hum that means 2 times 10 bits, and I had only 18 pins left once the Ultrasound SDK board connected. That means.. quite obviously, 2 times 9 bits.
So now I can run at twice the speed of the transfer, so a bit over 20Msps, over this time 9 bits. Or by oversampling, that could be 10Msps at 10 bits. With more flexibility !
Win.
At least for the moment, now I need to order the PCBs and some components.
-> BOM is around 30$ components, 5$ OSHPark PCB.. that's quite neat.
Again, been playing with Jupyter to test this news ADC: https://github.com/kelu124/bomanz/blob/master/ADC08200/20170430-ADC08200-FirstAcqs.ipynb
This yielded interesting results: here's for example a small emulation of a ultrasound signal at 2MHz: we do see something
Not bad! The oscilloscope gives the same results:
You will note that I have put an offset of around 650mV to the signal so that it fits within the range of the ADC. The FFT domain does not look bad neither:
So I'm quite happy with it! Again see the full notebook at https://github.com/kelu124/bomanz/blob/master/ADC08200/20170430-ADC08200-FirstAcqs.ipynb
FFT for this 1MHz signal.
It seems to work! A quick and inexpensive 10Msps DAQ for the pi - the first of the project series. When fed with a 2V (out of 3.3V range), the Pi Zero receives a 10 points period.
The waveform is encoded on 6 bits, that's not much but at least the proof that the pi0 can handle it is here =)
And one needs to test if they work !
are here too =)
Yeay! The PCBs for 20Msps+ ADCs is here =)
Congratulations on being one of the Internet of Useful Things Hackaday Prize Finalists!
Love the ultrasound imaging project, maybe this could also give you some interesting ideas on ADC https://hackaday.io/project/9877-analog2pi
could this be used to create crazy oversampled oscillators for audio/music/synth purposes?
Hey @Damon the ADC would need to be replaced by a DAC, but I assume indeed you'd be able to synthethise signals up to 10Msps =)
kelu124
julien
spencer
Jeremy
Jenny List
Hi @kelu124, for a sensor application I am testing, I would need a AD converter with a sampling rate of 20Msps connected to a RTU for doing live data processing. I know that RasberryPi cannot achieve that; have you developed your project further, do you know systems which can achieve that (without going to NI products), or would you be interested in cooperate to build one?