|
peepsalot posted:OK, answering my own question, but apparently two of the pins can actually be used as sense wires to allow the PSU to compensate for voltage sag when drawing high current (max rated 150W, or 4.16A per power/ground pair) The extra pins might not handle the extra load, but the mandatory ground on pin 8 would let the card check for a suitable supply. Once upon a time, it was popular to have multiple independent +12V rails in an ATX power supply. I assume this simplified engineering compared to a single rail having to support 400-600W. Handwaving a bit, but if pin 8 connects to a sense pin on, say, the power management IC, and it's also tied to one of the +12V rails via a pull up resistor, the card could check if it's on a rail that can handle 150W before powering up that massive GPU or whatever.
|
#
?
Feb 28, 2014 21:05
|
|
|
|
| # ? Jun 8, 2024 00:03 |
|
|
quote:Hakko FX-888 Wow, didn't exist such a unanimous and resolute suggestion. Thanks, guys.
|
|
#
?
Feb 28, 2014 23:25
|
|
|
PSU manufacturers attach three pins to 12V by convention, not by spec. One of them is a sense line. That's what you're missing. 6 pin: two +12V power lines, two common returns, one sense line (to com), one NC. 8 pin: three +12V power lines, three common returns, two sense lines. (So you can detect if they've plugged a 6P cable into an 8P socket.) e:f,b
|
|
#
?
Mar 1, 2014 03:29
|
|
|
Yo, I want to build a thing ~for a project~ and I'm wondering if it's something I can do myself, or maybe I should find someone experienced to design it (how much would this even cost?). I want a board with a bunch of PCI-E sockets. The data/clock lines don't need to be connected, but the sockets need to be fully powered and a microcontroller on the board needs the ability to turn power to each individual socket off/on. I guess there needs to be some kind of power supply on board too. I've never really designed a PCB before, but I have plenty of experience with breadboards and soldering and programming stuff.
|
|
#
?
Mar 1, 2014 16:35
|
|
|
e: screw that, not gonna do it. 
Bad Munki fucked around with this message at 17:11 on Mar 1, 2014 |
|
#
?
Mar 1, 2014 17:09
|
|
|
Curve tracing frontend for analog scopes: http://www.kerrywong.com/2013/09/09/yet-another-semiconductor-vi-curve-tracer-ii/ It looks so goddamn awesome. I don't have an analog scope here, but I still want to build some sort of equivalent device. Maybe performing all of the curve tracing and measuring before dumping the information over USB?
|
|
#
?
Mar 1, 2014 18:42
|
|
|
Illusive gently caress Man posted:Yo, I want to build a thing ~for a project~ and I'm wondering if it's something I can do myself, or maybe I should find someone experienced to design it (how much would this even cost?). I want a board with a bunch of PCI-E sockets. The data/clock lines don't need to be connected, but the sockets need to be fully powered and a microcontroller on the board needs the ability to turn power to each individual socket off/on. I guess there needs to be some kind of power supply on board too. So all you need it to do is distribute power? That should be easy enough, sounds like all you need is a/a set of regulators, a microcontroller, and a bunch of FETs to switch the power. How much power will the things be drawing, and how many different voltage levels do you need?
|
|
#
?
Mar 1, 2014 19:35
|
|
|
Illusive gently caress Man posted:Yo, I want to build a thing ~for a project~ and I'm wondering if it's something I can do myself, or maybe I should find someone experienced to design it (how much would this even cost?). I want a board with a bunch of PCI-E sockets. The data/clock lines don't need to be connected, but the sockets need to be fully powered and a microcontroller on the board needs the ability to turn power to each individual socket off/on. I guess there needs to be some kind of power supply on board too. Wow, Robie. That brings back memories. Anyway, seems like a pretty reasonable project. A real quick look suggests you need 12V and 3.3V? So add up the power you need (how much per PCI-E socket on each one?). I suggest you find DC-DC converter modules and use integrated load switches (rather than discrete fets) to switch/limit/fuse power (although that depends on whether you do a PCB or breadboard, discrete fets are easier to breadboard). I may go looking myself but if you post the power requirements I'd spec out the power supplies and power switches for you.
|
|
#
?
Mar 1, 2014 21:48
|
|
|
Illusive gently caress Man posted:Yo, I want to build a thing ~for a project~ and I'm wondering if it's something I can do myself, or maybe I should find someone experienced to design it (how much would this even cost?). I want a board with a bunch of PCI-E sockets. The data/clock lines don't need to be connected, but the sockets need to be fully powered and a microcontroller on the board needs the ability to turn power to each individual socket off/on. I guess there needs to be some kind of power supply on board too. A really simple way of doing this and avoiding deep-diving into power / switching supplies would be to find off-the-shelf monolithic/module-based converters with enable/shutdown pins, and then run those to the MCU. You can probably look at Tyco/amphenol for the actual sockets themselves. Keep in mind though that some cards will behave in different/interesting ways without a valid link.
|
|
#
?
Mar 2, 2014 09:03
|
|
|
The PCIe spec has a listing for how much current the 3.3V, 3.3V standby, and 12V rails take per card. Looks like Linear Tech has reposted the requirements in one of their design notes: http://cds.linear.com/docs/en/design-note/dn346f.pdf Designing a switcher or three isn't the end of the world (especially considering you won't need to worry about preventing noise pickup to nearby data lines), but a newbie power designer will very often run into weird design errors. Sticking with modules is safer unless you're okay with potentially doing a second debug spin of your board. On the plus side I'm sure a few of us here would be willing to help debug if you went the discrete route. 
|
|
#
?
Mar 3, 2014 23:42
|
|
|
movax posted:Keep in mind though that some cards will behave in different/interesting ways without a valid link. Yeah I'm having a hard time imagining the application. Do the cards have some side channel? What's the use in powering them if you're not going to speak?
|
|
#
?
Mar 4, 2014 00:03
|
|
|
JawnV6 posted:Yeah I'm having a hard time imagining the application. Do the cards have some side channel? What's the use in powering them if you're not going to speak? I'm betting it's buttcoin related
|
|
#
?
Mar 4, 2014 00:12
|
|
|
Even then you need to get some data across the link. As is the cards will boot up and switch to the compliance pattern when they don't see anyone else out there.
|
|
#
?
Mar 4, 2014 00:16
|
|
|
JawnV6 posted:Even then you need to get some data across the link. As is the cards will boot up and switch to the compliance pattern when they don't see anyone else out there. Yeah, I was thinking buttcoins also and then realized that you sort of need PCIe links for them to be useful  Another thing though, you probably want to bus PERST# to all slots, and certainly run that to the microcontroller as well.
|
|
#
?
Mar 4, 2014 00:58
|
|
|
I wouldn't be surprised if the new "coming soon" BFL ButtThrasher 9000 was so awesome it couldn't actually use the data connection and instead communicates via USB, but still has to be in a pci-e slot. e: wouldn't, not would. Bad Munki fucked around with this message at 01:15 on Mar 4, 2014 |
|
#
?
Mar 4, 2014 01:08
|
|
|
Bad Munki posted:I would be surprised if the new "coming soon" BFL ButtThrasher 9000 was so awesome it couldn't actually use the data connection and instead communicates via USB, but still has to be in a pci-e slot. https://products.butterflylabs.com/homepage-new-products/300-gh-bitcoin-mining-card.html Connectivity posted:USB 2.0 - Bitcoin Mining Monarch cards can be used as an external computer peripheral and chained via USB hub. In this mode it can be controlled via an Android host or standard Linux or Windows computer.  I really think they're trying to say "gen1" but... why...
|
|
#
?
Mar 4, 2014 01:13
|
|
|
No buttcoins involved, it's more networking related. My boss decided to hire a student to do it for cheap (and the kid sounds pretty smart/experienced) so I'm off the hook and back to writing code. Edit: oh yeah and we thought about using the perst pin, but we're not entirely sure the card will listen to it / properly reset. Will probably do some tests when we have something wired up. Illusive Fuck Man fucked around with this message at 02:28 on Mar 4, 2014 |
|
#
?
Mar 4, 2014 02:25
|
|
|
Illusive gently caress Man posted:No buttcoins involved I'm not sure if I'm more or less disappointed. I'm really curious what it's specifically about now, though.
|
|
#
?
Mar 4, 2014 15:12
|
|
|
I'm thinking of using twisted pair cable in a bit of an unusual way. I want to drive a 50ohm load with a ~30v *analog* signal with frequency content around 1 - 15 MHz. I was looking at this stuff: https://www.cablewholesale.com/products/bulk-cable/cat-5/product-10x6-622nh.php 24awg is rated to carry ~3.5A so I should be fine there. In terms of the frequency response, it looks like I'll see about ~2db loss (see 16MHz on datasheet; my length is about 20m). Characteristic impedance is 100ohms which is kind of annoying but not a huge deal. Any considerations I may have missed here? I like the above cable as its cheap, but if anyone has any suggestions for other manufacturers feel free to suggest. Aurium posted:First I found this on amazon. But one of the reviews said that it still has a coating, just much thinner than normal. He still had to scrape it off, which didn't really save him any work. Cool, that seems reasonable.
|
|
#
?
Mar 5, 2014 01:18
|
|
|
Soldered a TSSOP the other day, It felt so  . Don't know if it actually worked though...
|
|
#
?
Mar 5, 2014 19:33
|
|
|
Cyril Sneer posted:I'm thinking of using twisted pair cable in a bit of an unusual way. I want to drive a 50ohm load with a ~30v *analog* signal with frequency content around 1 - 15 MHz. I was looking at this stuff: Why not just use 50ohm coax or similar to avoid the impedance discontinuity/keep it simple? Cost, creativity, experimentation? What is the slew rate/rise time of the signal like?
|
|
#
?
Mar 5, 2014 19:54
|
|
|
Get a transformer to match the impedance at the load if you want it work well. At 15 MHz and 20 meters you're well within transmission line theory, and you'd get 10% of the power you put in to the line in return, which might be an ok amount, but RG-58 would be the natural choice for that range and frequency. RG-174 is thinner and lossier, but probably not as bad as TP. In principle you should be able to get a 50 ohm line with two parallel 100 ohm lines, maybe give that a shot? E: Or if you have a common ground/shield and the load is possible to split, then terminating it as 2x50 ohm with the middle connected to ground could give better performance since some of the differential energy could end up as single-ended due to mismatches in line length. longview fucked around with this message at 20:30 on Mar 5, 2014 |
|
#
?
Mar 5, 2014 20:23
|
|
|
Is there a voltage difference across an ideal current source? I don't think there is but I don't know how else to solve this circuit. I'm given both currents. I'm just missing a piece however I set it up with Mesh-Current or Node-Voltage analysis. It's having two current sources that's throwing me off.
|
|
#
?
Mar 5, 2014 21:10
|
|
|
The Royal Scrub posted:Is there a voltage difference across an ideal current source? I don't think there is but I don't know how else to solve this circuit. The impedance across an ideal current source is infinity. As Wikipedia's reminded me, that implies that the voltage across the current source is variable based on what you have hooked up to it: 0V in a short circuit situation, and increasing in voltage the more load you put on it. Edit: This seems like this would be a lot easier to calculate if you turned those current sources to equivalent voltage sources. Iris of Ether fucked around with this message at 21:27 on Mar 5, 2014 |
|
#
?
Mar 5, 2014 21:24
|
|
|
The Royal Scrub posted:Is there a voltage difference across an ideal current source? I don't think there is but I don't know how else to solve this circuit. Have you been taught superposition yet?
|
|
#
?
Mar 5, 2014 21:34
|
|
|
Iris of Ether posted:Edit: This seems like this would be a lot easier to calculate if you turned those current sources to equivalent voltage sources. Hmm I thought about that but didn't think I could do that to both current sources. Otto Skorzeny posted:Have you been taught superposition yet? No but I know it's in the book. I'll take a look. Edit- Node-Voltage and 2 KCL equations made the problem really easy. God I suck at this, thanks for the help. 
Coldstone Cream-my-pants fucked around with this message at 22:55 on Mar 5, 2014 |
|
#
?
Mar 5, 2014 21:48
|
|
|
The Royal Scrub posted:Is there a voltage difference across an ideal current source? I don't think there is but I don't know how else to solve this circuit. A voltage source varries current in order to regulate voltage -shorts are bad (infinite current), opens are fine (no current) -a voltage source always has a current limit -a capacitor is a type of voltage source A current source varries voltage in order to regulate current -opens are bad (infinite voltage), shorts are fine (no voltage) -a current source always has a voltage limit -an inductor is a type of current source I find it helpful to illustrate that everything in the voltage world has a parallel in the current world. So yes a current source has a voltage across it - it has whatever voltage results in the current it's trying to drive.
|
|
#
?
Mar 5, 2014 23:20
|
|
|
movax posted:Why not just use 50ohm coax or similar to avoid the impedance discontinuity/keep it simple? Cost, creativity, experimentation? Experimentation. The 50ohm load is actually a sensor in a remote, noisy location. Contrary to popular belief, coax is not the be-all-end-all of RFI protection; its shielding is not perfect and ultimately the sensor lines necessarily must leave the shielding providing a nice little antenna. I was thinking that a balanced load driven through a shielded twisted pair line off a balanced source might actually result in better noise immunity.
|
|
#
?
Mar 6, 2014 03:05
|
|
|
Cyril Sneer posted:Experimentation. The 50ohm load is actually a sensor in a remote, noisy location. Contrary to popular belief, coax is not the be-all-end-all of RFI protection; its shielding is not perfect and ultimately the sensor lines necessarily must leave the shielding providing a nice little antenna. I was thinking that a balanced load driven through a shielded twisted pair line off a balanced source might actually result in better noise immunity. The first thing any EMC book tells you is: shielding is useless if you half-rear end it, so if you want to remove the shield, do it inside a shielded box with a 360* shield connection, if you remove the shielding then it's not hard to imagine the shielding is removed. Coax is problematic in installations where ground loops can be formed, since it can interfere with the utility signal, the shielding itself is quite good, at least 40 dB rejection over frequency to outside RF for single shielded and >90 dB for triple and semi rigid. Try doing the 2x100 Ohm line I suggested above if you're into experimentation, I'm actually pretty curious to see how that will perform. If the impedance is matched then a balanced load/balanced driver and a shielded cable is probably going to outperform coax for RFI and signal integrity in challenging conditions. If you get problems with integrity, try terminating it to shield as well.
|
|
#
?
Mar 6, 2014 07:45
|
|
|
I have a question about an AC circuit where I think I have to use voltage division in the phasor domain. Only I know voltage division isn't defined, so I'm not sure how to do something like (120∠0)[(10∠90)/(23.7∠35.75)] I found an example where the magnitudes are multiplied/divided and the angles are added/subtracted. Seems weird but I'll go with it and hope it gets me somewhere. Coldstone Cream-my-pants fucked around with this message at 03:00 on Mar 7, 2014 |
|
#
?
Mar 7, 2014 02:45
|
|
|
The Royal Scrub posted:I have a question about an AC circuit where I think I have to use voltage division in the phasor domain. Only I know voltage division isn't defined, so I'm not sure how to do something like (120∠0)[(10∠90)/(23.7∠35.75)] Nobody really uses phasors in the real world (except maybe industrial engineers for power distribution). Working with complex numbers is generally easier. So just convert them to complex numbers, do the algebra like you would with typical resistive dividers (except with complex resistance), then convert the answer back to polar/phasor if you want.
|
|
#
?
Mar 7, 2014 04:29
|
|
|
The Royal Scrub posted:I have a question about an AC circuit where I think I have to use voltage division in the phasor domain. Only I know voltage division isn't defined, so I'm not sure how to do something like (120∠0)[(10∠90)/(23.7∠35.75)] According to wikipedia (because I can't remember because I only had to do this for one class) its just like multiplication except divide the magnitudes and subtract the angles I do remember that phasor is the easy one for mult/div and hard for plus/minus
|
|
#
?
Mar 7, 2014 05:08
|
|
|
Anybody wanna critique a design I've got going here? This is just for a personal project, an LED lamp (the same one that came up last page) with some extra stuff (capacitive sense switch, max brightness adjustment, fade on/off, multiple brightness levels, stuff like that.) This is a board with a grid of LEDs on one side and some control/supply stuff on the other. So starting from left to right: There's a three-pin connector. This will provide 12VDC and ground from a wall wart, and the center pin will go to a bit of foil elsewhere which will act as a capitative sense switch. P2 and C4 are are for the capacitive sense switch. That's a 2MΩ pot, supposedly 1MΩ is about the right value for the behavior I'm looking for, but I would like to be able to tune the sensitivity without updating hardware/software, so I figured using a trimpot as a variable resistor that covered more than the range I need should do fine. That's why one pin on the trimpot is NC. C4 is supposed to improve the stability of the sense. This is the example I'm using for the switch: http://playground.arduino.cc//Main/CapacitiveSensor?from=Main.CapSense Next up is the controller, an attiny45. It will be used to handle the capacitive sense switch, as well as control brightness via pwm. The max brightness is set using another trimpot, p1, acting as a voltage divider from the 5v rail. At the moment, I'm leaving the two resistors on either side (r11, r12) as 0Ω jumpers, but I wanted to have the option of adjusting the range of output. I've also put pads on the attiny45 for direct access to the pins should I find a need to reprogram it in place or something. After that is a transistor array that will be used to control the actual leds. All eight of the channels will be controlled by the same pwm output of the attiny45, so they're all tied together. Then there are ten channels of leds, in order to keep it at a convenient 12V (3.2V per led). I doubled up the channels on either end in order to get ten channels out of eight, but the current limitations should be fine there. Up at the top, again left to right, there's a pfet to provide reverse polarity protection, so that if the connector for this thing gets hooked up backwards, it won't get damaged. The output of that goes both directly to the leds and the transistor array, and to a 5v linear regulator that will be used to power the attiny45. The caps attached to the regulator are straight out of the example in the data sheet. Aside from the image above, the whole thing is available on github along with a board layout, but I figured I'd leave that for later. The whole project is on github if that's useful, or if anyone wants it: https://github.com/gshort/lamp
|
|
#
?
Mar 8, 2014 19:33
|
|
|
Bad Munki posted:Anybody wanna critique a design I've got going here? This is just for a personal project, an LED lamp (the same one that came up last page) with some extra stuff (capacitive sense switch, max brightness adjustment, fade on/off, multiple brightness levels, stuff like that.) This is a board with a grid of LEDs on one side and some control/supply stuff on the other. Two small thoughts 1) Since you have the pfet already throw a soft start on it with an R and C. Tie the gate to ground through a resistor and add a cap between 12V and the gate. When 12V is first applied the cap will be at 0V so it will pull the gate up and the pfet will remain off. The resistor will pull the gate down at a controlled rate and make a nice soft turn on. Set the RC for 10-100mS. Besides being good practice you could also add a larger on-board bypass cap without causing a spark when you plug in the power connector. 2) The led brightness might not be terribly uniform if that matters. Going by spec LED's have a pretty wide forward voltage range - over the 36 ohm resistor that's going to be a fairly wide current/brightness range.
|
|
#
?
Mar 8, 2014 21:26
|
|
|
I think the leds I got are binned according to their various characteristics, so that should help. And they will be behind a diffuser (a bit of frosted acrylic) so that should help to alleviate most of the differences. Hopefully. Thanks for the tip on the pfet, I'll see about adding that in.
|
|
#
?
Mar 8, 2014 21:28
|
|
|
Remember to throw some decoupling caps in there. Why are you using a darlington array instead of just a single power MOSFET? And what's going on with those ATTiny pins?
|
|
#
?
Mar 8, 2014 23:13
|
|
|
You mean the X's? Those are just blank pads that will be on the board. They're larger then the teensy-weensy pins on the attiny itself, so if I ever want to hook up to the thing for reprogramming or whatnot, it'll be easier. Like so: As for the array vs. the power mosfet...I don't have an answer. It's possible at some point the channels were going to be individually controlled? I don't recall anymore. 
|
|
#
?
Mar 8, 2014 23:20
|
|
|
Bad Munki posted:You mean the X's? Those are just blank pads that will be on the board. They're larger then the teensy-weensy pins on the attiny itself, so if I ever want to hook up to the thing for reprogramming or whatnot, it'll be easier. Like so: You were  ing about getting the fets into a single package. It looked like the design is well within the uln2803's ratings - is there a reason to not use a single monolithic part?
Delta-Wye fucked around with this message at 01:22 on Mar 9, 2014 |
|
#
?
Mar 9, 2014 00:53
|
|
|
Delta-Wye posted:You were :spergy: ing This is undoubtedly true.
|
|
#
?
Mar 9, 2014 00:59
|
|
|
|
| # ? Jun 8, 2024 00:03 |
|
|
What are you doing for LEDs, anyway? I'm planning on making a lamp in the near future, too. I like the idea of capacitive brightness adjustment.
|
|
#
?
Mar 9, 2014 03:09
|
|