Pretty good weekend effort

Thread Starter

joeyd999

Joined Jun 6, 2011
5,234
The whole point of stabilizing the temperature -- aside from characterizing the temperature sensor performance -- is to characterize the noise performance of the transducer output. Here is a 15 minute run of unfiltered data from the transducer:

Selection_010.png

The drift and undulations are a function of the mC changes in temperature. This is not inherent to the transducer -- but inherent to the nature of the test. In its proper operation, the transducer is actually immune to temperature changes (within its specification).

The total noise is about 1.7 counts RMS. At the level shown (about 811,430 counts), this represents about 0.0002 percent of the signal amplitude.

And this is only the high end A/D. The low end A/D drops the noise down to less than 0.1 counts when it's operating within its range.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,234
Here you can clearly see the correlation of drift to temperature (again, nature of the test, not the transducer):

Selection_011.png

The graph also gives you an idea as to why I had to pin down the temperature so tight in order to get good noise estimates.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,234
Wow! I think I found a bug in my code.

I changed the temperature output resolution from 1 mC to 10uC (xx.xxxxx) and look how the data comes out:

Selection_012.png

Notice the individual steps? There should be about 170 steps per mC, but it is clear they are resolving to a far fewer number of steps. This makes me think I've got an error in my ADC -> temperature conversion routine. If it were working correctly, it should look very much like a noisy analog waveform.

I've never noticed it because I never tried to look at uC before!

Seems I got work to do.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,234
There is another temperature sensor on board. It uses different processing. This is what it looks like (again, 10 uC resolution):

Selection_013.png
 

wayneh

Joined Sep 9, 2010
17,496
Here you can clearly see the correlation of drift to temperature (again, nature of the test, not the transducer):

View attachment 155112

The graph also gives you an idea as to why I had to pin down the temperature so tight in order to get good noise estimates.
I’d plot one against the other and drop the time axis. Why? Because you can! But seriously, time should not be important per se, right?
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,234
I’d plot one against the other and drop the time axis. Why? Because you can! But seriously, time should not be important per se, right?
It's nice to know someone is paying attention.

I've been at this for 12 straight hours today, most of the time watching the two tracks in parallel in 0.001C resolution. They looked exactly the same.

As a last task tonight, I changed the resolution to 10uC and immediately they looked different, causing me to jump to a quick conclusion without thinking.

Watching the grass grow for hours at a time must have an effect on one's head.

I completely forgot that, while my base data rate is 64 SPS, the data rate for that temperature sensor is only 7.5 SPS. That is why it looks like there are distinct levels as opposed to the second temperature sensor that is, in fact, running at 64 SPS.

In fact, the rounding algorithm was hiding some of the noise. Adding precision made the noise (and low SPS) more apparent.

But everything is still OK. 0.001C is good enough.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,234
By the way, @wayneh, I've been doing all this work on Kst. It's pretty awsome, and I haven't even dug into a portion of the features it has to offer.

Nice thing: it works with static or dynamic data files with any number of columns. So, if you've already got data stashed away somewhere, it's a breeze to do a lot of quick and dirty analysis. Or even not so dirty.

I'm doing a long data run tonight. 64 SPS in the Igloo until I get back to work in the morning. I'm excited to see the results.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,234
You do realize that sort of temp control has value in the market, right? I don't know much about the lab equipment business and what their SOTA is, but I know they charge dearly for their products.
Oh, yes!

Unfortunately, I chose early in my career to work in a very small niche market.

If I had chosen industrial or scientific controls, I'd be a very rich man today.

Those markets have huge, long-term players. I have no "in", and no one in that industry will take a small -- and new -- guy like me seriously.

With that said, I do get calls from customers that have "discovered" some of my products that are applicable to what they do. Names like Lockheed, NASA -- and a lot of universities. The first thing they usually tell me on the call is "Damn. Never used a product so good."
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,234
This is what the "active" portion of the controller looks like. It mounts into a wall of the Igloo.

There's also a thermistor that is installed in the assembly, but it is not shown in these photos.

IMG_20180627_115012.jpg IMG_20180627_115022.jpg
 

wayneh

Joined Sep 9, 2010
17,496
Those markets have huge, long-term players. I have no "in", and no one in that industry will take a small -- and new -- guy like me seriously.
A little Marketing could go a long way. ;) I won’t say it’s easy, but there are customers for your products and Marketing could help them find you.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,234
A little Marketing could go a long way.
Yeah. If I had a dollar for every time I've been told this (and tried it), I'd also be a very rich man.

It really isn't that easy, especially when these markets tend to be dominated by Good Ol' Boys clubs, where if you are not long-time established member, you don't get the time of day.

This is especially true on the brick-and-mortar distribution side (where most volume commercial sales take place).

I've been doing what I do for a little less than a hundred years. Been there, done that.
 

cmartinez

Joined Jan 17, 2007
8,218
It's DONE! ... my own "pretty good weekend effort" (which actually took almost 4 months of full work) is fully operational and working perfectly! ... and all thanks to the PIC10LF322 ... which only consumes 0.031 µA @ 31 KHz ... :)

Now it's MY turn to celebrate ...

AmpleBowedAoudad-max-1mb.gif
 
Top