Jump to content
esaj

Headache

Recommended Posts

16 minutes ago, esaj said:

Second table finally in place (and missing some shelves, still). A lot of time's been spent just dismantling and moving stuff around to get space.

Table brought in in pieces:

Nice sturdy work spaces!

 

17 minutes ago, esaj said:

Love that japanese saw, btw, if you work with wood, try one sometime. After getting used to it, it kicks ass compared to western saws:

handmade-ryoba-02.jpg

 

I just had a  few Amish gentlemen work on my gazebo and pole barn and they used similar saws. Power tools are forbidden. 

  • Upvote 1

Share this post


Link to post
Share on other sites

Ahh I see you took my suggestion for the angle braces. I like it! I still think you need some type of cover for them though. I would hate to nick a finger on that corner. Or a knee.

  • Upvote 1

Share this post


Link to post
Share on other sites
Posted (edited)
On 8/7/2018 at 12:12 AM, Esper said:

Ahh I see you took my suggestion for the angle braces. I like it! I still think you need some type of cover for them though. I would hate to nick a finger on that corner. Or a knee.

Yeah, the ends of the steel bars are filed down, but it wouldn't hurt to add something to "cushion the blow" in case I manage to smack some body part there... probably I could cut some cushions from old styrofoam pieces or whatever.

I had to take some break from building the tables', as my sister was visiting for a few days, so I'll get back on track in the next few days. I did get a few things done before that though. A good while back, I actually bought an "electrical distribution center" with DIN-rails, so it did come to some use here:

lAwXTeY.jpg

2 x 2-rail 6A automatic fuses and a couple of switching ("impulse") power sources (12V & 24V, both 150W). The fuses are twin-rail, because I can't know which way around I'm putting the socket in (ie. either rail could be phase or null, since the socket we use here do not force you to place it in any particular position).

UuGHIxB.jpg

Most of the wiring done and cables tied. 

wER51eu.jpg

I placed the cabinet near the sockets under the 160cm table. In hindsight, I should have gotten DIN-railed devices with both automatic fuses + fault current protections (they were out of stock when I got the fuses), I separately bought ground fault circuit interrupter (GFCI), or Residual Current Device (RCD) later on that you can put on the sockets.

RDokIBU.jpg

Left some open space there for the power sources to cool down.

1sFgiKR.jpg

Still ways to go, I hope to have enough time in the next couple of days to finish the last desk, some shelves and clean out the mess, I've got a couple of projects to do...

The 160cm table on the left also has wiring done for 230V mains, 12V and 24V mains (running under the first shelf and through a breaker box with rocker switches to turn on/off the switchers), some LED-lighting from the 12V rail (and an ugly power distribution board for 12/24V power, that I plan to replace once I get the CNC back up and running), although I probably have to add more, and a separate 100VA isolation mains transformer bolted to one of the posts and going to an ungrounded extension cord, not seen in the picture. There's also a (grounded) extension cord feeding from the (ungrounded) distribution box, might need to add another one once I can mill out proper distribution boards.

Also quickly made some small shelves to hold the sample books from the leftovers of the old tables:

wDBkB4N.jpg

Edited by esaj
  • Like 2

Share this post


Link to post
Share on other sites
Posted (edited)

Finally on the finish line with this, although there's still junk to organize and the steel pipe ends are completely unprotected  :P 

RCKvCd5.jpg

There's only so much stuff that you can fit into about 6m(about 65 square feet) of usable floorspace (the entire room is about 3x3 meters = 9m, but there are cabinets and the door on one wall, which take away about 1 meter so that the doors can still be opened). I had to get rid of one drawer to fit the floor rack and CNC under the tables, but now I've got a lot more desk space for working and could fit the 3D-printer here.

 

kRtj7Ea.jpg

tiMaH7R.jpg

The "measurement/testing" table with oscilloscope, bench multimeter, power sources etc. I might still add one more shelf on top.

qLz6NIH.jpg

Oei18uw.jpg

Tons of components, drawers, CNC-table and the floor rack fit under the desks, except for the cardboard box housing the component bags and one desktop computer, all the drawers, the rack and CNC table are on wheels so they're easy to move around when needed.

He3nRsQ.jpg

"Manufacturing/assembly" table, holding all sorts of tools, space for soldering and assembly, CNC/3D-printer -control computer and 3D printer.

nx0XKuv.jpg

Main computer desk, which also holds all the component drawers & misc crap on top  ;)

So, I had to do some compromises, but I increased the desktop area from about 2.5 square meters to about 3.8 square meters. The biggest improvement is the space for soldering/assembly and testing things, and not having to move crap around the room all the time to be able to move  :P

Edited by esaj
  • Like 3

Share this post


Link to post
Share on other sites
12 minutes ago, Hunka Hunka Burning Love said:

:popcorn:  Maybe a couple of these monitor wall mounts would free up some more counter space.

https://www.amazon.ca/AmazonBasics-Heavy-Duty-Articulating-12-inch-39-inch/dp/B01KBEOGZ4

Maybe, but not much and the "main" computer's already on the desk behind the monitors (since I can't fit it under the desk) :D 

At one point I was thinking of getting a large curved 4K TV to replace the dual monitors, but then I'd have to switch this machine to Linux too, because Windows window manager sucks. it's easy to throw, say, KiCAD schema-display to one monitor in full screen and PCB layout to another (they're interlinked, clicking a component on either will center the other to the same component, which is really useful and saves a ton of time when there's lots of components on the board):

PEFP3Fs.png

With single large display, I'd always have to be stretching the windows so that they don't overlap and would accidentally make one fullscreen, then have to resize and align the windows again etc.

  • Like 1

Share this post


Link to post
Share on other sites

Awesome work space! Well organized! So glad you included some essentials.   ;)

43283680874_16fa08524e_b.jpg

 

  • Like 1
  • Upvote 1

Share this post


Link to post
Share on other sites
Posted (edited)

Back to business.

I've got some components made for the load earlier (temperature-sensitive PWM-controller for turning main fan on when things heat up, distribution board for ATX-power supply), that aren't really interesting. In the meanwhile, a friend of mine mentioned he'd like to start using switching-mode power supplies for his analog synthesizers, as they are cheaper, light weight and much more efficient than bulky large linear supplies. The problem, as usual with switchers, is noise. For digital circuits, the noise is not a problem, as they usually work with a large hysteresis / "dead band" between the voltages representing 0 and 1, but for analog circuits, the noise is a problem, especially when you need high accuracy and very low noise. With a synthesizer using something like 1V per octave (for representing frequencies of notes, about 83.3mV per half-step) for example, a typical 1% ripple noise in 12V line is already 120mVpp. Other types of analog circuits also suffer, the lower voltage differences you need to "distinguish"/use for whatever reason, the more of a problem it becomes.

There are switcher-based power modules for synthesizer available, but they're surprisingly expensive and usually very simple. For example:

https://www.befaco.org/en/lunch-bus/

https://www.befaco.org/docs/LunchBus/Lunchbus_sch_1.0.pdf

As a parts-kit, this power supply module costs a little under 100€ with shipping, and 120€ fully assembled (but the shipping is then free). It's not really complicated, and the parts cost is about 30-35€ (including the board, the Mean Well DKE switcher-module is the most expensive part, I bought one for about 21€). Befaco's an "open source hardware" -shop, so they show the full schematics and allow you to build your own from scratch if you want (which is really fair), and probably these are really low volume products, which explains the high prices (they can't buy the parts in bulk, and they need to run the business, if after taxes and other expenses etc. you get a profit of something like 10-30€ per module, you need to sell a truckload of those just to keep in business ;)).

This isn't much more than a distribution board for the DKE-module though. And still needs a separate external power supply, such as a laptop charger. There are lots of different power supply modules out there, some good, some bad, some expensive, some less expensive. Most of them still let significant spikes through, although usually the modules have their own bypass-caps, possibly more power supply filters and op-amps and such reject at least some of the ripple in the voltage.

But, naturally, I wanted to build something that's better, and possibly even cheaper vs. buying a ready-made module or parts kit. The load is supposed to be powered by an ATX-power supply, which usually have that around 120mVpp ripple. With high current modules, one millivolt can mean 10mA, so I also have use for attenuating the ripple at least on the +12V line (not sure of -12V line yet).

Linear regulators are "nice" in the way that they at least somewhat attenuate voltage ripple by themselves, usually there's a graph for PSRR (power supply rejection ratio) SVR (Supply voltage rejection), ripple rejection or such (the names vary) in the datasheet that shows how much (under specific conditions, the ratios can change with input/output voltage differences, currents, temperature etc.) of the ripple they attenuate. Many of the old "jelly bean" regulators, like LM317's or LM78xx's don't attenuate much past 10kHz or so, as they were designed at the time when switching mode supplies were rare, and usually the interesting frequencies were mains line frequencies and maybe a couple of harmonics above them (like 50/60Hz, 100/120Hz etc). More modern LDO (low dropout) regulators commonly have better attenuations at higher frequencies, and I managed to find some that show graphs up to 1, 10 or even 100MHz. Scouring through the datasheets, I picked a few of these (on the cheap end, of course) to test around with.

LM2941 is a relatively cheap (0.99€ / piece with 24% VAT at 10+ pieces in TME) 5-20V / 1A LDO with surprisingly good PSRR (in the datasheets at least):

RXCZ0G5.png

 

That's almost a straight line! There's a clear dip somewhere around 100-300kHz though, which also happens to be the quite typical switching frequency of cheaper switchers, like computer ATX's for example (usually around 150kHz). But still, the rejection ratio should be about 65dB. Also note that this is measured with 10mA output current, typically the rejection gets worse with higher currents. 

Another one I got was LD1085, which is clearly meant as a post-switcher regulator, since the graph starts at 10kHz (or then it's an error in the datasheet, it happens ;)):

M2190Wa.png

Notice also that here it's measured with full maximum output current (3A). The rejection stays above 70dB up to until about 3MHz and then rolls off, still about 40dB at 100MHz (again assuming that they really mean kHz). 

Of course there's more to selecting regulators than just the rejection ratios, but for this use, it was a pretty dominant parameter. I also got a couple more models than the above mentioned, but haven't played around with those (of some I have only one piece, as they were "expensive", like 4-5€ per piece), and also some for negative regulation to test out circuits for my friends' synths, as he needs +-12V dual supply (I don't really have that much need for it, probably I'll use the +-12V dual supply for testing op-amps using the +-15V DKE and regulators to drop it to +-12V with filtering, as a "side product" my friend gets the design for his synths).

A regulator won't work "alone", usually the rejection is dependent (at least partially) on the output bypass capacitors, together with the output impedance of the regulator and the capacitive reactance of the bypass capacitor(s), they form a RC-low pass filter (well, "ZC" if you like). The reason the attenuation doesn't go on "forever" (like you would assume from the basic equations for RC or LC-filters) is that real world components have "parasitic" properties (for example, an inductor will have some stray capacitance) which works "against" the ideal function at high frequencies. Imagine that an inductor's reactance goes higher with the frequency, but at the same time, there's some capacitance in the physical inductor. At high enough frequencies, the capacitive reactance of the inductor becomes low, and the higher frequencies actually go through this parasitic reactance, bypassing the inductance.

I tried the regulators on a breadboard, and managed to drop the 120mVpp ripple spikes from the ATX down to about 20mVpp at 1A output, just with the LM2941 regulator and a couple of bypass capacitors (didn't write down the values though):

Ad7Dtdx.png

The yellow line is the input from the ATX, red is output measured over a bypass capacitor. 1x probe attenuation, 20MHz bandwidth limit, AC coupling. Do note that they have different scales (10mV per div for red, 50mV per div for yellow).

With lower currents, I could get down to around 10mVpp at times.

It's not easy to measure things like these with breadboard, as there's a ground loop formed by the ground clip of the probe (can't really use ground spring there) and the breadboard itself has all sorts of bad connections and stray inductances and capacitances. Still not bad just for the regulator and a couple of chinese noname electrolytic caps (as a comparison, I earlier tried with some "jelly bean" LM1117's or whatever on a milled board, three in series dropping something like 12V->9V->7V->5V and the ripple didn't really attenuate much at all).

With the LD1085, I got down to about 13mVpp at best with 1A current (using some low ESR-polymers too, can't use those with the LM2941, as it's the kind of design which requires a certain MINIMUM amount of ESR in the bypass caps, or otherwise it may become unstable), but didn't take scope shots or write down the components (again). Also, most of the time I couldn't get below about 20mVpp on the breadboard with higher currents, maybe because of the poor measurement capabilities with the breadboard, or just stray parasitics and bad (other) components, as I used just some noname-chinese cheapo caps.

Late last night I started to design a quick prototype board, and today I finished the design & milled a prototype board.

vxQV6CM.png

Nothing complicated, this is just for testing anyway.

 

ZN7GcNt.png

I didn't try really hard with the layout, just made sure that I have at least a little space around things and didn't create any ground loops.

The board was laid out so that I can test both LM2941 and LD1085 with it (not at the same time of course) ;). Do note that the R1 & R2 values must be the other way around for LM2941 (whoops), otherwise the output voltage won't be 10.2V but around 3.2V  :P  

I added possibilities for using THT-electrolytics, SMD polymers and MLCCs, as well as ferrite beads and an optional inductor for further filtering in the output. The LM2941 is finicky with the output capacitor ESR, thus the possibility of adding resistors in front of MLCCs in the output. After milling the board, I populated it first for use with the LM2941. At this point, the input side had a few MLCCs, two 100uF / 35V polymers (no idea of ESR, unknown chinese), L was left out (0R resistor there) and the output side had only two electrolytics (unknown chinese brand, one measured 435uF with 0.3ohm ESR, the other had 428uF and 0.31ohm ESR). Both ferrites were in place (unknown chinese 1206's with supposedly 1k ohm impedance at 100MHz).

J4I7zeX.png

I had the resistors the wrong way around, so the output voltage was about 3.2V instead of 10.2V. The measurements are done with again with AC-coupling, 1x probe attenuation and 20MHz bandwidth limit, this time using a ground spring to measure over the last bypass capacitor. Again note that the scales are different (2mV per div vs. 50mV per div). It's a bit finicky to measure with the ground spring, as even holding from the "right" spot on the probe can give out extra noise, and if (and when) the probe tip or spring moves accidentally while you hold it there adjusting the scope, the signal can go really haywire ;)

At one point I thought that I had hit the jackpot and the spikes were attenuated completely:

NChjmSI.png

No such luck, it turns out that the noise floor of the scope is about 800µV (0.8mV), and the probe tip wasn't properly touching the board, while the ground spring was :D  On the other hand, nice to know that I can't even get below 800µV in measuring, so if I really manage to get the noise that low, I know there's not much point to continue (unless I get a better scope :P).

I then added two 10uF MLCC's (Samsung X5R, 25V) with 1.3ohm front-resistors to stay within the allowed minimum ESR and replaced the 0-ohm resistor with a Ferrocore HPI0630 8.2µH inductor (around 60mOhm resistance, Irms is around 3.6A) to get some further post-filtering going on. The post-regulator circuit looks about like this:

80EgOv1.png

Now I could get (at best) the ripple down to below 3mVpp:

axMMdxn.png

Not bad. For further reduction, likely better / more filters would be needed, but I'll be next desoldering the LM2941 and seeing if I can get better attenuation just by replacing it with LD1085, maybe I can use low ESR polymers on the output-side with it (which I'm not actually sure I can do if it does have a minimum ESR-requirement, on second check the datasheets don't say anything about that :D). At least it should be possible to bypass one of the adjust resistors to ground with that (the LM2941 datasheet explicitly forbids bypassing either adjustment resistor), that should help with the spikes. Then there's still a few more regulators to test, some which specifically state no need for minimum ESR for stability, which helps as when you add more caps with low ESR you get much lower impedance for higher frequencies, but they're on the more expensive side (plus with the DKE, I'll have to test the negative regulators also at some point, which usually have far worse high frequency PSRR on their own).

 

 

 

 

 

Edited by esaj
  • Like 1
  • Upvote 1

Share this post


Link to post
Share on other sites
On 8/12/2018 at 1:18 PM, esaj said:

With single large display, I'd always have to be stretching the windows so that they don't overlap and would accidentally make one fullscreen, then have to resize and align the windows again etc.

There is an option to do that automatically. Right click on your taskbar and click on 'show windows side by side' and if you have windows 10, there is a button to show multiple desktops, so that way if you want to just open up a browser and search something without having multiple programs open (which would affect the 'show side by side' feature) you can click on that button. I feel like you may already know this, but in case you didn't, I hope this helps.

  • Upvote 1

Share this post


Link to post
Share on other sites

×