|
Phone posting but let me try to draw the options I was thinking of: This one to limit sound level to about .7v, which may be too quiet:  This one to get 2x the zener voltage, so you can choose a zener diode for the appropriate voltage level. 
|
#
¿
Jun 11, 2020 16:41
|
|
|
|
| # ¿ May 9, 2024 09:01 |
|
|
You know now that I think about those circuits they work entirely by loading the amp. That's probably not great because if the amp wins they burn out and you sudden get a burst of loudness anyway. Maybe drop about 22 ohms in series with this circuit's non-ground input so it's dropped in a resistor instead of the amp. e: like this so the amp sees 22+your headphones ohms normally and 24 ohms when the limit hits. 
Stack Machine fucked around with this message at 16:58 on Jun 11, 2020 |
|
#
¿
Jun 11, 2020 16:48
|
|
|
Dawncloack posted:This is also a pretty cool idea, I'll start doing the math soon. Thanks! Not a link but search terms in quotes: the problem is good audio amplifiers are all "closed-loop amplifiers" They use negative feedback to keep their output at a given voltage. If we just "clamp" their output to a given value, they'll drive the maximum current they can source into that load to try to get it at whatever voltage they're trying to achieve. Put another way, the "small signal output impedance" of a good audio amp is very very small. So we add a series resistor in series with the amp and before our clamp so the amp gets to drive what it wants to and the extra power is burned up in our resistor.
|
|
#
¿
Jun 11, 2020 18:29
|
|
|
Put another way, the voltage gain of an audio amplifier is fixed, so if we just clamp its output like in the figure on the left it'll drive as much current as it can to get that output to the voltage gain times the input. With our resistor, like in sketch on the right, we attenuate the signal in ordinary operation but we also limit the current from the amplifier through the clamp, in this totally made-up case to about e: 59mA. I realized while I was walking the dog that there IS a speaker in this system too. Stack Machine fucked around with this message at 20:53 on Jun 11, 2020 |
|
#
¿
Jun 11, 2020 20:42
|
|
|
You can get ethernet jacks with "integrated magnetics" now too, just in case board area is too tight to have a separate transformer.
|
|
#
¿
Jun 12, 2020 01:58
|
|
|
Dawncloack posted:
The 220 is between the input and ground (i.e. in parallel with the rest of the circuit). I think it's there to set the maximum resistance of the whole thing to 220 ohms, while the 22 ohm in series that I suggested is there to set a minimum resistance of 22 ohms. The reason for having a max resistance is a little less clear-cut and has more to do with keeping distortion low but the minimum resistance is more important, since it keeps the amplifier from turning the circuit between itself and the speaker into vapor.
|
|
#
¿
Jun 12, 2020 14:24
|
|
|
Spatial posted:I've got a few odd valued ceramic capacitors in my design. I want to make the BOM simpler/cheaper by using smaller caps in parallel, summing up their capacitance values. They all have the same dialectric and voltage rating so it should be okay in theory right? In fact even better since the ESR is lower I think??? Those are odd-valued? Check out the wikipedia article on the E series of preferred values. 2.2 and 3.3 are both on E6, the series for components with 20% tolerance. Maybe things from E3 (the 40% tolerance series) are cheaper in MLCC land? 2.2 is still on that. Unless it's for electrical (esl/esr) reasons adding <10% of the value in parallel with what are presumably 20% tolerance caps looks silly. It's like topping off a jug containing 0.8-1.2 liters with a tablespoon.
|
|
#
¿
Jun 13, 2020 19:25
|
|
|
Spatial posted:The values aren't unusual, I meant they're odd ones out in the BOM. It's just easier/cheaper to build the board with fewer unique parts.  that makes a lot more sense.One thing you can do is compute new tolerances for your aggregate devices and see if they're OK. Like a "22uF" made of 2 10uF caps will have a minimum value of 16uF with 20% components and a maximum value of 24uF and so you could claim it behaved like a "22uF" cap with 27% tolerance. Or you could get tighter-tolerance caps for the values you're doubling/tripling up. 2 10uFs at 10% is like a 22uF at 14%, etc.
|
|
#
¿
Jun 13, 2020 20:59
|
|
|
|
|
#
¿
Jun 15, 2020 21:39
|
|
|
KnifeWrench posted:I've come to learn that the "convention" for UARTs is confusing and so commonly misunderstood that you basically need to check the schematic. This makes that review all the funnier. There's no labeling of "tx" and "rx" here that's entirely wrong by every (generous) interpretation so the fact that they're completely shaken by tx/rx being swapped is some good pedantry porn no matter whether the board uses "yours" or "mine" labels.
|
|
#
¿
Jun 16, 2020 00:22
|
|
|
It looks a little bit like a 2-pin Molex micro-fit, but that has a 3.0mm pitch. All of those standard motherboard power connectors since atx are some form of micro-fit. There's also a nano-fit in 2.5mm but I don't think those have the angular style keys these have.
|
|
#
¿
Jun 19, 2020 05:41
|
|
|
I don't know about the stuff sagebrush uses but the odd spool of wire wrap wire I have is silver plated and teflon insulated. Unlike untwisted telephone wire which melts back like mad when it gets warm but is practically free if you don't value time spent slicing open cables and using a drill to untwist the pairs.
|
|
#
¿
Jun 21, 2020 01:28
|
|
|
Sagebrush posted:I have an LCD I'm trying to work with and I need to generate -15v from +5v. My favorite inverting topology is the inverting buck-boost. Something like this TI part:  That part or something similar could be configured to give you -15V. At 10mA and a duty cycle of 25% the inductor will only be handling about 30mA so it'll be cheap too. At those current levels you could also look into some sort of charge pump. Most of the LCD modules I've used have included charge pumps to drive the display and either integrated the capacitors or required the user provide a few capacitors. My work and play has been entirely on switched inductor supplies though so I know basically nothing about what's available in charge pump land. Stack Machine fucked around with this message at 06:42 on Jun 21, 2020 |
|
#
¿
Jun 21, 2020 06:15
|
|
|
Here have an inductor for that for about 12 cents in quantity 10. At 220uH, the current ripple would be around 12mA or so peak to peak so the max inductor current would be 36mA. Plenty of margin. I realize cheap passives aren't news but I have a ton of small valued, low energy caps and low-power resistors but every inductor in my lab has a core the size of my thumb so this seems really neat to me.
|
|
#
¿
Jun 21, 2020 06:41
|
|
|
I've always wanted to try to get a really pro look by buying an extruded aluminum case and having faceplate/back panels laser cut in 2-tone acrylic for highly legible etched lettering and holes for switches/connectors/screws. The only machining operation required for this would maybe be counter-sinking for the screws. I think it could look more polished than a 3D print but I've never seen it done.
|
|
#
¿
Jun 21, 2020 22:27
|
|
|
Daang. So this is a sine wave oscillator set up as an inverter with a PWM controller providing closed loop control by flipping the power to said inverter on and off? That's one hell of a circut.
|
|
#
¿
Jun 22, 2020 21:50
|
|
|
Keep in mind when playing with trade-offs in these kinds of circuits that shrinking the output cap is always an option if the loop gets faster but less stable. So you lose some stability by moving the feedback network but you gain some speed (i.e. loop gain at higher frequencies) too so then you shrink the cap to bring back that stability and the ripple is maybe the same but the cap is smaller. If you're lucky, that equates to cheaper components or less board real-estate.
|
|
#
¿
Jun 23, 2020 15:55
|
|
|
General_Failure posted:Just asking for thoughts on this. I replaced the right analog stick on a very battle weary PS4 controller. Did a crap job. Took me way too long to get the old solder off. According to a PC joystick testing program the right stick is always pressed. It isn't. Button still clicks. Presumably this is a normally-open switch. One possibility is that its default "open" state of this part is still too low resistance. There's not much you can do about that. Possibly adding a series resistor would help but unless it is right on the threshold it doesn't look good. Another is that the solder flux is slightly conductive and now all over your board. You can help that by cleaning the board with isopropanol.
|
|
#
¿
Jun 24, 2020 06:25
|
|
|
Ambrose Burnside posted:If I went with the 555, would I need to make any other changes to the circuit beyond a higher-voltage power supply and altering the proscribed pin-out to suit? Drop a 1k-10k resistor in series with the LED for the 555 circuit. The 555 is not designed to drive an LED and will likely burn it up at higher input voltages.
|
|
#
¿
Jun 26, 2020 03:29
|
|
|
Technicians back when wire wrap/vectorboard was big used to follow the schematic or netlist marking off connections as they went and then go back through and double check as they buzzed it out with a meter. Obvs less of a thing now with surface mount and custom pcbs where that's all done by DRC/LVS but for breadboard there's no shame in making a netlist and drawing checkboxes on it.
|
|
#
¿
Jun 26, 2020 19:17
|
|
|
When I say netlist I mean specifically a list where each circuit net is a section and each component pin it connects to is an entry in that. Like GND: R1-1 U10-14 U1-6 ... Then you can rake your meter probe over all the pins and make sure it only beeps on those pins. It's time consuming but doing those checks means you don't have to be perfect the first time every time or make smoke. Past like a dozen nets that seems to be crucial. But also PCBs from the likes of 3pcb are really cheap now so if you want to move to that, your software can do those checks for you.
|
|
#
¿
Jun 26, 2020 19:22
|
|
|
Those were meant to be component reference designators (i.e. R1-1 is the first pin of the first resistor, U1-14 is the 14th pin of the first IC, etc.) not coordinates. So it's easier to compare against the schematic to make sure the list is correct. Also easier to modify.
|
|
#
¿
Jun 26, 2020 20:18
|
|
|
Ambrose Burnside posted:yet another change of subject: i stripped an old microwave for parts- Ambrose Burnside posted:2. what is this part? I think that's a bimetal thermal switch to cut the magnetron power if it starts getting hot where the food is.
|
|
#
¿
Jun 26, 2020 21:14
|
|
|
Maybe I'm just cavalier but I wouldn't write off the MOT for anybody at least comfortable playing with 120V. A MOT with the secondary coil cut off and new secondary windings installed is a good basis for basically any mains-connected transformer you could want, from power for an audio amplifier to an isolation transformer to power for tube circuits. Just don't plug it in with the original secondary installed unless you really have a load in mind for like 2kV at a goddamned amp.
|
|
#
¿
Jun 27, 2020 06:08
|
|
|
E: bad and factually incorrect post, but linear interpolation is good and should be used. It's big cost, however, are the multiplies and adds I didn't even mention here. Stack Machine fucked around with this message at 01:36 on Jul 1, 2020 |
|
#
¿
Jul 1, 2020 00:29
|
|
|
I'm sorry I realize looking at my post that it is too terse to be of any value and actually is not really applicable. Let me try again. A 256-entry LUT is not what you want for this, probably. The function you link to looks like it is intended as a form of dynamic range compression. The whole reason for doing this is to preserve information from the LSBs for very "small" values while keeping range for larger values, much like floating point. If you throw away your LSBs and then use a lookup table, you will still lose information from your LSBs. You will be using one of 256 input values to select one of 256 code words. This function is monotonically increasing and non-linear. This means all 256 code words can't possibly appear in the output of your LUT and some must necessarily repeat. You will lose the detail at small values the function is designed to preserve and also throw away some of the data from just using your upper 8 bits directly! My original comment mentioned linear interpolation, complete with an incorrect definition. (I've just had knee surgery and I think I'm still a little anaesthesia-woozy) The idea there is you take the values a = lut(floor(x*N)) and b = lut(ceil(x*N + epsilon)) and compute a*(ceil(x*N + epsilon)/N-x) + b*(x - floor(x*N)/N), effectively drawing a line between the two LUT values and selecting a point from along that line. Not only can you get by with fewer LUT entries, even with a very basic approximation, as long as your multiplies are 16-bit, you still fulfill the function's original goal of dynamic range compression. E: "epsilon" here in the Calc 1 sense of a very very small number. I.e. ceil(1.0 + epsilon) is 2. Stack Machine fucked around with this message at 01:41 on Jul 1, 2020 |
|
#
¿
Jul 1, 2020 01:31
|
|
|
Logarithmic tuning capacitors with spiral-shaped blades have been made before if you want some inspiration. They occasionally pop up at hamfests and the like.
|
|
#
¿
Jul 2, 2020 22:16
|
|
|
Kerbtree posted:Can someone please stupid check me before I let the smoke out? Yes. The only thing you are trying to avoid is connecting your PC ground to your virtual ground through the USB/audio cables, so breaking the ground connection on either side should be enough.
|
|
#
¿
Jul 3, 2020 00:53
|
|
|
ante posted:No, that's what he asked for, but we don't know the application, so we don't know how critical accuracy is. I suggested throwing out the LSBs earlier. I made a post earlier to this effect and I don't want to beat a dead horse before the OP replies, but this is processing samples from an ADC and it's intended to be set up in a way that preserves resolution for small signals; it's meant to be dynamic range compression. Throwing away your LSBs before performing dynamic range compression defeats the purpose.
|
|
#
¿
Jul 3, 2020 01:04
|
|
|
Ok, I'm going to rear back and strike this dead horse hard because I think maybe just asserting that I don't like your idea without trying to communicate why is just rude, and if there's one thing I can't stand, it's the idea that I might come across as rude on the something awful dot com forums. I don't know what the exact function in the plot is, but let's assume it's f(x) = tanh(x/8192) where x is the signed 16-bit sample value, from -32768 to 32767. If we plot that, it looks kind of close to the linked plot:  Now, if we plot a histogram of the number of input codes that map to each of the 256 output codes, we get this (Y axis zoomed in because the middle values are what I want to focus on here):  So we get 60-80 codes mapping to each 1 code at the lowest signal levels. This means we could discard a few LSBs, since even at the highest-gain portion of the curve we have all of the values of the lower 5 or 6 bits mapping to the same output code. So that gets us to maybe 10 bits. What happens if we go beyond that? Here's the same histogram, but we quantized the input to 1 of 256 values (i.e. knocked off 8 LSBs) first:  Gross! A lot of the output codes are now useless and small signals get a lot of additional quantization noise. Because I was curious I decided to check how many effective bits were left in the output code, or how many of the 256 codes can actually be reached. 117 of them; less than half. You could represent that with 6.9 bits.
|
|
#
¿
Jul 3, 2020 01:46
|
|
|
babyeatingpsychopath posted:What tools are you using? This is a skill I want in my toolbox. We can move this into some other thread that's more appropriate, if you wish. I'll post about it here for now because making little plots, computing error bounds, etc. comes up a lot in electronics and having tools available for these tasks is important. I made these plots and histograms with GNU Octave, which is an open-source Matlab clone. I use Octave quite a bit for generating plots for little reports and emails and the like as frequently it's easier to make a point with a few plots than a load of text. I'm not sure I'd recommend Octave specifically vs something like R or some plotting lib for Python or some other tool, but I use it because Matlab was part of my undergraduate curriculum and I never really got it out of my system. It's also an industry standard in EE and especially DSP. The actual code I used was something like: code:
|
|
#
¿
Jul 3, 2020 20:01
|
|
|
Ambrose Burnside posted:Trimmer capacitor build questions: I think you're talking about the fringe effect. As long as the dimensions of the overlapping region between both plates are large compared to the distance between them, this is not a significant contributor and you can just pretend that you have a parallel plate capacitor with the area of the overlapping region. If you want to work through the calculations, you can find an upper bound assuming you will have a fringe effect around the entire perimeter of the overlapping region. Ambrose Burnside posted:2) If there are multiple materials with differing dielectric constants between plates, how do you determine the final effective constant? If the material is just stacked up in layers, you can treat it as a set of capacitors in series. The fact that there is not an actual conductor between them doesn't matter and the calculations are exactly the same. Just like connecting two identical parallel plate capacitors in series is equivalent to doubling the distance between the plates. If the material is side-by-side, same argument but the equivalent capacitor is parallel. If it's some complex geometry, break it down into series and parallel blocks. Ambrose Burnside posted:3) Assuming a multi-rotor/stator design using thin plates: What happens if the dielectric constant between sets of plates varies? with a standard rotor trimcap: each effective pair of plates consists of the nearest conductive opposed surfaces, which means that any one rotor/stator made of homogenous conductor is serving as two capacitor plates simultaneously, each plate consisting of either opposed plate surface, right? What happens if you made the plates from an asymmetrically-dialectric plate, like protoboard with copper cladding on one face and FR4 on the other- would anything change if you arranged the plates "head to tail" vs "head to head, tail to tail", with the former having the same constant among all pairs and the latter (assumably) having alternating low- and high-constant pairs? Again, you should consider the equivalent circuit of series and parallel capacitors that form your cap. Your standard tuning capacitor (or MLCC for that matter) is a bunch of identical capacitors in parallel. In the case you describe in which those units are not the same, you will have a few different sub-types of capacitor and your total capacitance will be the sum of the capacitances of the sub-units.
|
|
#
¿
Jul 4, 2020 00:54
|
|
|
Ambrose Burnside posted:The specific thing that's throwing me here is imagining a "piston" trimcap vs the more usual rotor type, where you have a rod-shaped rotor entering or retracting from a tubular stator. If I'm visualizing things correctly: with a plate rotor sort, yeah, it's clear how the actual overlapping regions are going to be equal no matter what. With a piston design, though, the element with the smaller effective radius will always have a smaller plate area than the larger element, and that imbalance is maintained no matter the rotor position. I think. If so, how do you square it- find the average of the two plate areas? The most accurate answer is that you have to start with basic equations and re-derive the capacitance for the coaxial arrangement instead of the parallel plate arrangement. What happens in practice, though, is that the gap between the piston and sleeve is made intentionally small compared to the radius of the piston or sleeve, meaning that the plate area is roughly the same and the geometry is "close enough" to the parallel plate arrangement that it can be modeled as such. Compare: http://hyperphysics.phy-astr.gsu.edu/hbase/electric/capcyl.html With: http://hyperphysics.phy-astr.gsu.edu/hbase/electric/pplate.html For situations where (b-a)/ln(b/a) is approximately b, these equations are equal. This happens when b and a are close in value and in these situations it doesn't matter which one you use since the difference in their surface areas is going to be much smaller than their surface areas. To provide a concrete example, say the radius of the piston's outer surface is 100mm and the radius of the inner surface of the sleeve is 101mm. 1/ln(101/100) is 100.5, which would be the "effective radius" for modeling as a parallel plate capacitor. This is very close to the average distance, but as the gap gets wider the approximation becomes less valid, 10/ln(110/100) is 104.92. Stack Machine fucked around with this message at 17:34 on Jul 4, 2020 |
|
#
¿
Jul 4, 2020 02:46
|
|
|
Ambrose Burnside posted:Neat, thanks. I’ve only ever run into the parallel plate model + formula so I didn’t realize other weirder capacitor geometries have their own derived equations to work from, but in retrospect it makes sense. What you're really talking about in a capacitor is the amount of work needed to move a unit of charge from one plate to another as a function of the amount of charge difference that already exists between the two plates, and all of the models with simple equations make assumptions so this is easy. The cylindrical capacitor link I sent, for example, assumes the cylinder is of infinite length. I had some errors in my algebra earlier so I corrected the earlier post just to avoid confusing anybody reading through the thread. You were right though; the average surface area is a very good approximation in the "piston" design. The following plot is the effective radius with a 100mm piston and 1 to 100mm gap:  And this is the error from using the average radius approximation vs the effective radius. It's never bad; it's only 4% when the outer radius is double the inner radius: 
|
|
#
¿
Jul 4, 2020 18:08
|
|
|
For any manufacturing process technology in the semiconductor there is always a set of supported, well-characterized devices. The design engineer is always free to draw their own polygons in the available layers but there are some devices that already have well characterized simulation models and you'd better believe those are the ones everybody picks unless they have a very good reason not to. Also this is almost certainly a "shuttle run" type arrangement like MOSIS, where everybody's part gets placed all together on one big mask and that mask gets stamped multiple times onto a few wafers and everybody gets a few wafers worth of their part. It's much cheaper since the expensive part (the mask set) is shared, but it also means that everybody has to run the same process. Nobody gets to choose which layers they get. That's OK though! Choosing layers is a cost saving thing where you say "none of my devices used the high-rho poly2 so I'm turning that one off" or "I bet I can do this part with only 4 metals instead of 5". What I'm saying is this is the real deal but the real deal isn't as wild west as PCBs where you can put literally any component ever made on any board made in any way. It's very dependent on the manufacturing process. This is, however, definitely for analog/power/RF. You can do anything you could want to do on a square cm of this old rear end process with an FPGA except make the non digital parts of a radio/power supply/op amp/ADC.
|
|
#
¿
Jul 7, 2020 20:58
|
|
|
Forseti posted:I think what we're really all interested in is whether or not we can put microscopic Dickbutt pictures in our unused space. I seriously doubt they're planning to do the layout themselves so yes. What is normally provided to the foundry is a GDS "stream file" containing a bunch of polygons with various attributes, much like a slightly less photoplotter-oriented version of the Gerber files used in PCB manufacture, but the layers are things like implants and oxide thickness and polysilicon. And the design rules are quite complex.
|
|
#
¿
Jul 7, 2020 21:18
|
|
|
This may have been posted here before but: https://micro.magnet.fsu.edu/creatures/index.html
|
|
#
¿
Jul 7, 2020 21:23
|
|
|
Sam Zeloof's work is in many ways cooler than google doing an open shuttle run because he gets around those expensive masks by using a DLP. That's the kind of poo poo I want to see. If we could do that industrially at competitive feature sizes we could have small production runs of wafer-scale ASICs. That's not practical now because reticles are just so loving expensive. I keep making this analogy but it could be a lot more like PCBs are now where you could just order 5 or 6.
|
|
#
¿
Jul 7, 2020 21:39
|
|
|
Ambrose Burnside posted:took another crack at designing myself a trimmer capacitor and i think i've got sth nifty, wanna run it past the brain trust for critique before I go and assemble it. These look rad. You can make your traces wider to kill parasitic inductance and that'll make your capacitor look more like an ideal capacitor at higher frequencies. You could even make your stator a solid rectangle instead of a triangle if you want to really minimize that. This mattering requires short wires to the rest of the circuit, so if the plan is to have it on a foot or 2 of hookup wire the last inches are less crucial.
|
|
#
¿
Jul 10, 2020 05:47
|
|
|
|
| # ¿ May 9, 2024 09:01 |
|
|
Dominoes posted:3: Polynomial, with 3 or 4 terms. This will probably be the best approach from accuracy, memory, and coding perspectives. If you do go this route, the one piece of advice I have is to factor your polynomial. This way you don't also have to do additional multiplies per term to find x^2, x^3, etc. Makes your code more compact and saves cycles. The compiler can't perform this optimization for you because order of operations affects rounding.
|
|
#
¿
Jul 13, 2020 02:10
|
|