Electronics > Projects, Designs, and Technical Stuff
EMF pickup from amplifier in I2C line causing glitches. (Now with scope trace!)
Starlord:
--- Quote from: dmills on July 05, 2016, 09:41:15 am ---Here https://e2e.ti.com/support/power_management/simple_switcher/w/simple_switcher_wiki/2243.understanding-measuring-and-reducing-output-voltage-ripple is a decent discussion of measuring high frequency stuff in switching power circuits. Note in particular the improvement when using a very short spring clip rather then the hopeless three inch wire with croc clip (The things are ok for strictly audio band measurements, but forget it for anything with RF involved).
--- End quote ---
I dunno... you're not making a very good argument here. I mean, sure, the measurements are more accurate, no argument from me there, but does it really make a difference in this application? If anything, it's merely makes the noise look a bit less severe. But the peaks are still clearly there. If i went to all the trouble to do this on my board, maybe I would see half the noise, but what's the point? It would still be causing glitches. I guess I could be a little more certain if my filtering would be sufficient to allow me to scrape by, but that's it.
--- Quote ---What is your board stackup, and is it the same as the TI eval board for the power amp? IME such things need to be on at least 4 layer boards to work well because you need a solid ground plane to stand any real chance at all.
--- End quote ---
It's a two layer. Another move to reduce costs. The ground plane is fairly unbroken around the amplifier though, and it's right next to the power input and some big capacitors. I also isolated the power plane from the rest of the circuit as best I could. I'm sure this isn't as good as a four layer board, but it is what it is.
Andy Watson:
--- Quote from: Starlord on July 05, 2016, 10:46:06 am ---I dunno... you're not making a very good argument here. I mean, sure, the measurements are more accurate, no argument from me there, but does it really make a difference in this application?
--- End quote ---
What is being said is that without proper grounding you could be chasing problems that do not exist or are insignificant. Here is a thread with some good links to videos about grounding scope probes:
https://www.eevblog.com/forum/beginners/current-measurement-70275/
dmills:
Seriously, pick up a cheap SDR and use it to make some frequency domain measurements, cost is only a few hundred quid (You can make reasonably satisfactory near field E and H field probes out of thin coax), and it will at least tell you where the glaring problems are (My bet, radiation from **all** the external cables, power, signal, you name it).
Hell, hire a spectrum analyser and calibrated Yagi, you can do much in an open field site for a day or twos time expenditure and a few hundred in hire costs, of course this might tell you that you have a serious problem, but precompliance work like that is much cheaper then failing at the test house for something you could have caught and fixed.
The FCC only got into the act after the early days of the home computer thing were over (Because they were causing so many problems), so those ribbon cable connected peripherals were ok at the time they were produced, but there was also much less use of the RF spectrum and it was mostly transmitters at far higher power levels then is common today, so the interference was somewhat less of an issue.
It is one thing to do a rush to market job when you know what you are doing, sure you might stuff up, but the basics will be there (Filter the switching converter, input power and output as appropriate), design with reasonable margins, you may get it wrong, but it will probably not be horribly wrong. It is quite another to go there without the understanding to read between the lines of a datasheet (And you always have to read between the lines on a datasheet, they are as much marketing as technical in many cases).
For example note that the app note for that power amp notes that if external DC power is used the cheapest way to keep radiation from the supply cables down may be to do a proper filter in the speaker output connections, major hint, also the presence of locations for snubber RC networks on the eval board, another not unreasonable hint. I would take the fact that all of the diagrams show an output filter as a sign that while it may be possible to pull off a ferrite bead filter, it is problematic to do well, doubly so if you don't have the tools to test it (Beads saturate, and they seldom give you a curve for current Vs inductance, unlike a proper inductor)....
You do things like filterless (Or bead inductor) when you are doing a run of 50,000 pieces or so and you can amortise the cost of the extra prototypes needed to get it right over the volume, for a run of a couple of hundred the extra prototyping pain will kill any savings.
The 10MHz thing is also not asking for a bead having a self resonance at 10MHz, but for a bead/capacitor network having a combined resonance below 10MHz, not the same thing at all.
Odds are if you can find a decent experimentally oriented ham radio club, you could get some useful hints there, and those guys understand this stuff (And may well have a spectrum analyser available).
In terms of bringing a product to market compliance testing is small beer, a year of engineering time (nothing much takes significantly less then this when you include everything) has a cost of at least £40K just in pay even if you ignore the hidden costs (at least 80K total is more like it), now add marketing, market research, moulded cases, production set up, all that stuff, and if you get any change out of £150K for a new product intro you are doing very well, £10K or so for the test house? Big deal.
Regards, Dan.
Starlord:
Okay, so I'm considering the possibility of foregoing the ferrite bead and putting an LC reconstruction filter or whatever you want to call it on there. Looks like a low pass filter to me.
But a 10uH inductor is too large, so I decided to do some calculations and see if I could get away with using a smaller one.
I found this calculator:
http://circuitcalculator.com/lcfilter.htm
So the first thing I did was I plugged in the values TI provided for the components. 10uH and .68uF.
So, L = .000010H and C = .00000068F. This gave me F = 61033 Hz and Z = 3.8348 ohms.
Okay, that seems to pass the sanity check. 61KHz would be in the range of what I might expect for a low pass filter that should leave the audio frequencies untouched.
So let's see what I can find on Digikey for shielded SMT inductors that are small (0.1" or less) and have high current handling capability:
http://www.digikey.com/short/3482r2
Okay, so looks like I've gotta design around something in the 1uH range. Let's plug that into our calculator, along with a 10uF cap. L = 0.000001H, C = 0.00001F.
Hm... F = 50329Hz. That's fine. And Z = 0.31623 ohms. Uh oh. My gut tells me that Z should be higher if it's to attenuate the noise as much as TI's suggested values.
Well, I can't really use an inductor twice the size. I wonder if this will be good enough? Time for another calculator. I couldn't find one which graphs the frequencies with just LC input, but I found an RLC one. I'm guessing for R I should input the DC resistance of the inductor. So I'll do that and see what I get.
Here's the calculator:
http://sim.okawa-denshi.jp/en/RLCtool.php
Let's try this one from Bourns. Fairly typical values:
http://www.digikey.com/product-detail/en/bourns-inc/SRP2512-1R0M/SRP2512-1R0MCT-ND/4876867
So I input 49m ohms, 1uH, and 10uF into the calculator and...
Fc = 50KHz. Okay that passes the sanity check.
Well, that's interesting. The bode diagram shows a relatively flat response up to around 50Khz where there is a peak, then after 60KHz there's a linear drop, and around 1MHz it reaches a 50db reduction.
How much is a 50db reduction?
http://www.satsig.net/lnb/db-calculator.htm
100,000x? Well damn, that's more than enough I think. And hell, 10K would probably be sufficient. I could reach a 40db reduction with a 0.5uH inductor and 10uF cap.
Thoughts on going this route instead of using the ferrite bead?
dmills:
Better, but you are still not, I think doing it quite right, far better to use something like ltspice (Free download, and pretty standard for this sort of thing) and remember to include the speaker as a load resistor.
With 10u you are going to get a lot of peaking at ~35KHz (~10dB) and the impedance seen by the amp falls off a cliff at the top of the audio band which may put the amp into current limit, I would be thinking less then 1u, maybe an order of magnitude less.
(1u + 100n) in each leg into 4 ohms comes out to -23dB @ 1MHz, -65dB @ 10MHz and more or less flat below 400k, your switcher is running at 1MHz, so there is no real need to put the filter too close to the audio band, especially if it puts a deep low impedance point where it can be expected to trip the power amp overcurrent protection.
Filters have two ends and you need to remember to keep an eye on the admittance at the input side or you can get into trouble, you essentially have to design the filter for a specific load impedance which means that you cannot just trade L for C (Keeping the corner frequency constant) without changing Q.
Regards, Dan.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version