Author Topic: FAST A>D conversion  (Read 5773 times)

0 Members and 1 Guest are viewing this topic.

Offline nctnico

  • Super Contributor
  • ***
  • Posts: 28429
  • Country: nl
    • NCT Developments
Re: FAST A>D conversion
« Reply #25 on: April 06, 2019, 06:33:01 pm »
to answer some questions,
no mutiplexing, just 30 leds wired direct to the binary>decimal convertor.
update rate of the leds would be the rate the data under the probe changes state - that's why the latch is between the dac and binary>decimal decoder.
now 20MHz is kind of worse case - most signals would be 4MHz or less - i picked 20 because it's kind of the max you would see a 74ALS series being run.
i know they can go higher but it's very rare.
I don't think this setup is very useful because your eyes are way slower compared to a 20MHz repetition rate. You really need to rethink this.
What if you would combine level and frequency into RGB leds. The number of LEDs on would be a measure for the amplitude and the color a measure for the repetition rate.
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Offline IDEngineer

  • Super Contributor
  • ***
  • Posts: 1950
  • Country: us
Re: FAST A>D conversion
« Reply #26 on: April 06, 2019, 06:54:26 pm »
no, that's a 2D display, i'm after a single row of 30led's. your on the right track though, imagine a VU meter with a 20MHz+ frontend.
I'm imagining that... and wondering why, and what you're going to do with it. Others have noted that your eyes don't have the temporal response to track changes in the 10's of MHz. So why do you need such a high A/D sample rate? Are you trying to display min/max voltages?

So far I'm missing the relationship between a relatively high bandwidth sample rate and a relatively low bandwidth user interface.
 

Offline stjTopic starter

  • Super Contributor
  • ***
  • Posts: 2195
  • Country: gb
Re: FAST A>D conversion
« Reply #27 on: April 07, 2019, 12:47:54 am »
the high bandwidth is needed to capture the state before it changes,
you dont need to update the leds at that rate because you only need to see the high and low signal levels to evaluate the quality of the signal.
so as i said, the led's can be latched at a few Hz.
 

Offline radiolistener

  • Super Contributor
  • ***
  • Posts: 4135
  • Country: 00
Re: FAST A>D conversion
« Reply #28 on: April 07, 2019, 01:19:58 am »
I think you can get some parallel ADC, such as AD9200 which was recommended above and connect it to the cheap Chinese eval board for CY7C68013A based logic analyzer:
https://www.aliexpress.com/item/EZ-USB-FX2LP-CY7C68013A-USB-logic-analyzer-I2C-serial-and-SPI-core-board-Source-Code-development/32851287266.html

it can handle 24 MHz sample rate for 8 lines or 12 MHz for 16 lines. And there is ready to use library to work with it even with no need to write program for CY7C68013A controller (this is a fast version of 8051-based architecture microcontroller with a high speed hardware implemented USB).
« Last Edit: April 07, 2019, 01:35:12 am by radiolistener »
 

Online joeqsmith

  • Super Contributor
  • ***
  • Posts: 12277
  • Country: us
Re: FAST A>D conversion
« Reply #29 on: April 07, 2019, 01:30:39 am »
So you want to know the voltage level before the signal starts to transition?   Normally, my projects are not that clean and I break out a scope to get a feel for the signal integrity.   

Are you planning to use a fixed threshold to trigger the device and just use some fixed precious time as the criteria?   

Maybe not the best example of messy signals but here I am playing with an eval board.   
https://www.eevblog.com/forum/microcontrollers/typical-speed-of-fpgas/msg1282290/#msg1282290

Offline stjTopic starter

  • Super Contributor
  • ***
  • Posts: 2195
  • Country: gb
Re: FAST A>D conversion
« Reply #30 on: April 07, 2019, 12:57:14 pm »
i have scopes,
and eval boards etc.

i'm looking to make something handheld so i dont have to keep looking down at chips and then looking up to a screen.
here is how it would function, it would grab a sample as fast as possible, purely to avoid sampling mutiple states in one shot,
then it would display the voltage as a simple led bar.

big bar = signal good, small bar = signal bad - probably from a failed chip output or blown internal pullup.
 

Offline IDEngineer

  • Super Contributor
  • ***
  • Posts: 1950
  • Country: us
Re: FAST A>D conversion
« Reply #31 on: April 07, 2019, 02:22:20 pm »
My inner Bob Pease is whispering that this should be done in analog, not digital. Maybe a pair of opamps, one each to detect and display the low and high voltages. Getting your 20MHz bandwidth is child's play in the analog domain. Total cost could be a couple dollars. You could still have the bargraph display you described, if that's even the optimal approach... lots of options.
 

Offline stjTopic starter

  • Super Contributor
  • ***
  • Posts: 2195
  • Country: gb
Re: FAST A>D conversion
« Reply #32 on: April 07, 2019, 06:25:30 pm »
maybe,
but i have over 30years of digital experience inc software.
my analog skills go about as far as selecting the right resistor to drive an led, and knowing how to get a clean output from a regulator!!
 :-+
 

Offline splin

  • Frequent Contributor
  • **
  • Posts: 999
  • Country: gb
Re: FAST A>D conversion
« Reply #33 on: April 07, 2019, 09:45:23 pm »
i have scopes,
and eval boards etc.

i'm looking to make something handheld so i dont have to keep looking down at chips and then looking up to a screen.
here is how it would function, it would grab a sample as fast as possible, purely to avoid sampling mutiple states in one shot,
then it would display the voltage as a simple led bar.

big bar = signal good, small bar = signal bad - probably from a failed chip output or blown internal pullup.

So you want a sampling rate of a few Hz with an acquisition time of 50ns or less to minimize the probability of sampling during a signal transition. You don't need a high speed ADC, just one with a fast(ish) sample and hold - or make your own sample and hold with an analog switch, a capacitor and a jellybean CMOS or FET input opamp.

I'd still suggest using a microcontroller as your solution is useless for signals which have a high or low mark/space ratio such as the write strobe of a RAM chip which might only be active for very short periods (a few hundred nano seconds perhaps) with relatively long off periods. The probability of sampling it in the active state woud be very low for sampling rates of a few Hz.

Using the $2 STM32L412 would be cheaper and simpler (H/Wwise). You would only need to use one of the ADCs running in 6 bit mode, configured for a sampling time of 2.5 clocks or 31.25ns. You could use your method by running the convertor a few times per second and calculate and output the bargraph driver signals on GPIOs but you could do much better by running both ADCs interleaved at 17.8MSPS and use the processing power to examine the signal. The capture and processing doesn't have to be continuous so you could capture almost 40,000 samples into a RAM buffer and then process it at leisure.

As a minimum you'd look for short pulses to ensure they get displayed. You could also look for overshoots/undershoots/oscillations in signal transitions and perhaps light additional LEDs to flag that it is worth further examination with a proper scope. Another possibility is to look for other anomolies such as runt pulses or glitches, and/or detecting a tri-stated signal level and displaying its voltage - even if it spends only short periods in the tri-state condition.

When updating the bar graph you could ramp up, or down, the level at a speed proportional to the signal transistion time (obviously within the limits of the 17.8MSPS  sampling rate). This could be helpful to detect overloaded signal lines or weak drivers.

Use two STM32L412s and interleave all 4 ADCs to get over 30MSPS for better signal timing resolution (but still subject to each ADC's 31.25ns minimum acquisition time).

Use the built-in USB peripheral to connect to a PC for uploading data or configuring thresholds, timings etc. relevant to the circuit being tested.

Whatever ADC you use you will almost certainly need to buffer the signal as high speed ADCs require low impedance drivers for accurate results. Without a buffer it could cause glitches to the signal as the ADC switches its sampling capacitor. The buffer would be very simple - just an opamp configured as a voltage follower. It would have to be reasonably fast to handle the signal frequency of course - say 50MHz BW or more, so not a cheapie LM324!
 

Offline IDEngineer

  • Super Contributor
  • ***
  • Posts: 1950
  • Country: us
Re: FAST A>D conversion
« Reply #34 on: April 08, 2019, 01:37:53 am »
the high bandwidth is needed to capture the state before it changes, you dont need to update the leds at that rate because you only need to see the high and low signal levels to evaluate the quality of the signal.
Despite your bargraph comments, it sounds like what you really care about is insuring that a logic low is "below a certain threshold voltage" and a logic high is "above a certain threshold voltage". That could be done with two discrete LED's.

For now, I'm sticking with my analog recommendation. Two analog comparators could do this in a single IC and deliver 20MHz without even breathing hard. No "missed levels because they happened between samples", either. And if you really want a digital output for some downstream purpose, the comparators can drive into digital circuitry... there are even comparators with open-collector/open-drain outputs for exactly that purpose. You can even have controllable hysteresis to reduce your noise susceptibility, basically for free.

This project reminds me of the guy who wanted to double the amplitude of an analog signal, and his plan was a full system of antialiasing filter, A/D, MCU, D/A, and output filter. That way the MCU firmware could multiply the samples by 2.  >:D
 

Offline stjTopic starter

  • Super Contributor
  • ***
  • Posts: 2195
  • Country: gb
Re: FAST A>D conversion
« Reply #35 on: April 08, 2019, 11:00:04 am »
I'd still suggest using a microcontroller as your solution is useless for signals which have a high or low mark/space ratio such as the write strobe of a RAM chip which might only be active for very short periods (a few hundred nano seconds perhaps) with relatively long off periods. The probability of sampling it in the active state woud be very low for sampling rates of a few Hz.

and there is the problem,
it's going to be checking clock,address,data and chip-select lines on old computers.
 

Online 2N3055

  • Super Contributor
  • ***
  • Posts: 7462
  • Country: hr
Re: FAST A>D conversion
« Reply #36 on: April 08, 2019, 11:11:58 am »
So you want min/max peak detector connected to LM3914 LED bar graph..
"Just hard work is not enough - it must be applied sensibly."
Dr. Richard W. Hamming
 

Online joeqsmith

  • Super Contributor
  • ***
  • Posts: 12277
  • Country: us
Re: FAST A>D conversion
« Reply #37 on: April 08, 2019, 12:23:30 pm »
So you want min/max peak detector connected to LM3914 LED bar graph..
This is what I keep thinking but does he really want to see the under and overshoot?   I think I just need to wait and see what he comes up with and how it is used.   

Online 2N3055

  • Super Contributor
  • ***
  • Posts: 7462
  • Country: hr
Re: FAST A>D conversion
« Reply #38 on: April 08, 2019, 12:28:47 pm »
So you want min/max peak detector connected to LM3914 LED bar graph..
This is what I keep thinking but does he really want to see the under and overshoot?   I think I just need to wait and see what he comes up with and how it is used.

I agree.. I kinda understand what he wants to measure. It's just I simply use scope for that..
"Just hard work is not enough - it must be applied sensibly."
Dr. Richard W. Hamming
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf