Author Topic: Product & Circuit Design Brainstorming  (Read 10515 times)

0 Members and 1 Guest are viewing this topic.

Yotson

• Newbie
• Posts: 1
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #25 on: March 07, 2017, 07:28:06 am »
Far from ideal as it gives a somewhat logarithmic result and has a temperature coefficient that has to be handled.  Both are talked about in the article. What about something like:

M@rcel

• Contributor
• Posts: 8
Re: Product & Circuit Design Brainstorming
« Reply #26 on: March 07, 2017, 10:36:35 am »
Perhaps a tip for USB isolation (see video at 29:00): Look into ICE08USBx from https://www.silanna.com/ (not affiliated in any way). The cheapest (I'm Dutch  ) product I could find that uses the ICE08USBC is found at http://hifimediy.com/high-speed-usb-isolator-480Mbps  (not affiliated).

They work very well. I use them for embedded design testing.

HTH
« Last Edit: March 07, 2017, 10:44:37 am by M@rcel »

xyzzy

• Contributor
• Posts: 16
Re: Product & Circuit Design Brainstorming
« Reply #27 on: March 09, 2017, 05:26:07 am »
Dave and David,

24 bits of dynamic range will get you from 10 nA to 167 mA.  A typical scope A/D is 8 bits.  What happens if you have 3 simultaneous analog outputs, each scaled by 256x relative to the previous output (ie. no gain, after one 256x gain stage, and after two 256x gain stages)?

For example, let's hook the 1x output to channel 1, 256x to channel 2, and 65,536x to channel 3 of your scope.  If your product has a 200 nA sleep current and a 10 mA wake current, then you'd read 78% of full scale on channel 3 during sleep, and 6% of full scale on channel 1 during wake (with channels 2 and 3 showing the full rail voltage of your opamp).  The best part is that, for those awkward measurements just beyond* a range threshold, you can adjust the vertical scale of each scope channel independently.

*http://www.catb.org/jargon/html/P/pessimal.html

BTW, I really enjoyed this video.  It's very educational to see the entire thought process (dead-ends included) vs just seeing an explanation of the final design.

FrankBuss

• Supporter
• Posts: 2321
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #28 on: March 09, 2017, 07:20:48 am »
24 bits of dynamic range will get you from 10 nA to 167 mA.  A typical scope A/D is 8 bits.  What happens if you have 3 simultaneous analog outputs, each scaled by 256x relative to the previous output (ie. no gain, after one 256x gain stage, and after two 256x gain stages)?

24 bits are a myth. If you carefully design the board, shield it, battery power it, use it at the lowest samplerate specified by the datasheet (higher samplerate means less usable bits, always read the fine print in the datasheets) etc., you might get 20 usable bits out of a 24 bit ADC. Once I used an AD7793 (24 bit ADC) on a 4 layer board, but my layout was not that good (e.g. no star-grounding, just one big ground plane, didn't know Dave's videos back then) and with a low samplerate and the internal voltage reference I got like 18 usable bits, and 20 bits with oversampling and averaging, which is pretty much what the datasheet says. With an external reference, 4 Hz samplerate and gain 1 the datasheet says 23 bits with RMS noise (meaning if you don't average, you don't get perfect values with this resolution), but 20 bits with peak to peak noise.
So Long, and Thanks for All the Fish
Electronics, hiking, retro-computing, electronic music etc.: https://www.youtube.com/c/FrankBussProgrammer

xyzzy

• Contributor
• Posts: 16
Re: Product & Circuit Design Brainstorming
« Reply #29 on: March 09, 2017, 05:41:43 pm »
24 bits are a myth. If you carefully design the board, shield it, battery power it, use it at the lowest samplerate specified by the datasheet (higher samplerate means less usable bits, always read the fine print in the datasheets) etc., you might get 20 usable bits out of a 24 bit ADC.

Yessir, I tend to agree.  My suggestion removes the ADC from the product entirely and relies on the ADC inside the user's oscilloscope instead.  By using 3 scope channels with 8 bits each and multiplying the signal by 256 and 256^2, respectively, in the analog domain, you can measure currents from nA to mA without autoranging.  You're effectively measuring three separate ranges in parallel, and when a lower range starts to clip, the next range up starts to provide useful information.  (This works with two-channel scopes as well; you simply omit whichever range isn't appropriate for your particular bi-modal distribution of currents.)  It's not a perfect solution, but I think it may be a cheap way to provide sufficient information in a useful form.

My only concern is if you can multiply an analog signal by 65536 (albeit in two stages) without introducing significant additional error.

bktemp

• Super Contributor
• Posts: 1623
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #30 on: March 09, 2017, 06:19:37 pm »
I would prefer having an ADC, because it allows a cheap digital isolation.
Being able to measure the current on the supply rail instead of the GND wire can be really useful in many situation.
If you still want an analogue output, 16bit DACs are quite cheap. If the ADC is good enough, you could add a dynamic range compression features like outputting the current waveform in a logarithmic scale instead of linear for viewing the waveform on a scope.

FrankBuss

• Supporter
• Posts: 2321
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #31 on: March 09, 2017, 06:22:46 pm »
I guess 24 bit dynamic range with analog components is also not possible, or gets really expensive, or slow, and if it is 20 bits, the last stage has only a dynamic range of 4 bits, which might be too noisy for many measurement applications.
So Long, and Thanks for All the Fish
Electronics, hiking, retro-computing, electronic music etc.: https://www.youtube.com/c/FrankBussProgrammer

hli

• Regular Contributor
• Posts: 205
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #32 on: March 09, 2017, 09:57:45 pm »
Linear has ADCs with 18 bits and 5 Msps, e.g. the LTC2385-18. You get 24bits at 2Msps with the LTC2380-24. Drawback: the 2385 comes only with LVDS interface.

f4eru

• Frequent Contributor
• Posts: 586
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #33 on: March 09, 2017, 11:01:19 pm »
Yessir, I tend to agree.  My suggestion removes the ADC from the product entirely and relies on the ADC inside the user's oscilloscope instead.  By using 3 scope channels with 8 bits each and multiplying the signal by 256 and 256^2, respectively, in the analog domain, you can measure currents from nA to mA without autoranging.

That doesn't work well. You need sme overlap, or you'll lose precision by getting some 2x steps, which is a very bad precision:
1,2,3.....,127,256,512,....

FrankBuss

• Supporter
• Posts: 2321
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #34 on: March 09, 2017, 11:43:20 pm »
Linear has ADCs with 18 bits and 5 Msps, e.g. the LTC2385-18. You get 24bits at 2Msps with the LTC2380-24. Drawback: the 2385 comes only with LVDS interface.

No, you don't get 24 bits at 2 Msps. Always read the fine print: 145 dB (=24 bit) dynamic range at 30.5 sps (still quite impressive), 100 dB SNR (=16.6 bits) at 1.5 Msps.
http://www.linear.com/product/LTC2380-24
So Long, and Thanks for All the Fish
Electronics, hiking, retro-computing, electronic music etc.: https://www.youtube.com/c/FrankBussProgrammer

hli

• Regular Contributor
• Posts: 205
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #35 on: March 10, 2017, 07:35:32 am »
No, you don't get 24 bits at 2 Msps. Always read the fine print: 145 dB (=24 bit) dynamic range at 30.5 sps (still quite impressive), 100 dB SNR (=16.6 bits) at 1.5 Msps.
http://www.linear.com/product/LTC2380-24
Ah, I was just looking at the overview table and not at the detail page. Dave actually wanted at least 18bits at more than 4Msps, so the LTC2385 was the better fit anyway.

stefanL

• Contributor
• Posts: 5
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #36 on: March 17, 2017, 10:22:44 pm »
A quick update on the RocketLogger: We have just made our project public, so you can now check out our design on our updated homepage.

The data includes:

Enjoy

We will also present our project at DATE 2017 in two weeks.
« Last Edit: March 17, 2017, 10:30:28 pm by stefanL »

ksrm

• Newbie
• Posts: 2
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #37 on: March 27, 2017, 08:59:59 pm »
That's a really interesting project, Stefan. I've also been thinking about how to do high dynamic range current measurements, but I thought I would need a much faster range switching time than the 1.4 us you have. Does it cause any problems in practice?

stefanL

• Contributor
• Posts: 5
• Country:
Re: Product & Circuit Design Brainstorming
« Reply #38 on: March 27, 2017, 09:34:15 pm »
That's a really interesting project, Stefan. I've also been thinking about how to do high dynamic range current measurements, but I thought I would need a much faster range switching time than the 1.4 us you have. Does it cause any problems in practice?

Thanks, it certainly was a lot of fun We never saw any problems related to the range switching with our own logger, while other we did with other equipment. The fact that the voltage drop is limited by the diodes helps a lot, also 1.4µs is not that long. That beign said, there are certainly situations, where it's still too slow and you just have to disable the low range.

I think it would be really hard to get even faster rangeswitching with our design. We ran into many issues around the same delay: Noise bandwidth, propagation delay and the RF filter step response. In that regard it might be interesting to look at a design that controls the input burden voltage (see f4eru's post).

Smf