Even someone as lazy as me has to get off his derriere for a friend. My friend and former colleague Neeraj Lal, from my NSC days, started a real company, Purilume, with a real product, the Orchid LED table lamp. Here’s a link to the Kickstarter campaign: https://www.kickstarter.com/projects/1415756116/orchid-lamps-experience-life-in-a-new-light
Way to go, Neeraj!!
I have been writing a technical training session on the LLC halfbridge converter, and one part that I believe is critical but doesn’t get the attention it merits is the reliability of the capacitor used in the resonant tank on the primary. Every LLC I’ve ever seen or designed used a 400VDC to 450VDC input bus, and every calculation I’ve ever done required a capacitance between 10 nF and 100 nF. For safety and engineering margin, that means that a DC voltage rating of around 500V or more is needed. The only capacitor type I know of that makes sense is a film capacitor. I discounted ceramic capacitors because of their capacitance loss vs. DC voltage.
From about half of max power to max output power, the current waveform in the resonant tank is nearly a pure sinusoid, so its RMS value is lower than a trapezoid wave or square wave of equal peak-peak amplitude – but even at a relatively low power such as the 120W of my test subject for this post the RMS current is around 850 mA. My test subject is the STEVAL-ILL052V1, an evalboard available from ST. For this post, I tested it with 300VDC applied to the main input, which is a boost PFC that outputs 450VDC. Here’s a scope capture of the current in the resonant tank and the voltage across the capacitor:
You can see why a ceramic capacitor with a widely-changing capacitance vs. DC voltage would be an absolute nightmare – there is 250VDC of voltage shift each cycle! The resonant tank capacitor is the B32652A0153K000 from Epcos, a 15 nF device rated to 1 kVDC or 250 VACrms. Its dimensions (this will be important later on) are: L x W x H = 18 x 5 x 10.5 mm. The datasheet notes «high pulse strength» and «for electronic ballasts and switch-mode power supplies». The temperature rating is from -55ºC to 110ºC, and I assume that means the sum of ambient temperature plus the temp rise from internal heating. What isn’t listed is an RMS current rating. The purpose of the post is to try and determine the reliability of this and other film caps based on information that is provided in the datasheets.
This family of film caps doesn’t provide ESR values either, but it does list dissipation factor, tan(sigma), at various frequencies. The resonant frequency of this test converter is almost exactly 100 kHz, and for family members less than 27 nF at 100 kHz, tan(sigma) is listed as 0.002. We can then calculate ESR:
ESR = (tan(sigma)) / (2*pi*fo*Cr), where «fo» is the resonant frequency and «Cr» is the resonant capacitance. ESR = 0.22ohm. That fits pretty well with the impedance curve shown farther on in the datasheet, if we assume that the minima represent ESR – I read about 150 mohm from that chart.
Power dissipation in the cap would then be somewhere between 0.85A^2 * 0.15ohm = 110 mW and 0.85A^2 * 0.22ohm = 160 mW. I set the converter up and duct-taped a thermocouple to Cr – it’s the grey blur in the middle on the right in the photo below:
T1 shows 37.3ºC, and T2, right at the bottom in middle, shows 22.7ºC in my lab. I took this photo after the system ran for about half an hour. The load voltage and current can be seen in the background. Empirically we now have a junction-to-ambient thermal resistance value range: deltaT = 14.6ºC at 160 mW, so thetaJA ranges from 91ºC/W (assuming 160 mW) to 133ºC/W (assuming 110 mW).
I found an app note from Vishay: https://www.google.es/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwjWv9Hi4Y_KAhUEVhoKHXoiCv8QFggyMAA&url=http%3A%2F%2Fwww.vishay.com%2Fdocs%2F26033%2Fgentechinfofilm.pdf&usg=AFQjCNFqu8LUJl4uVqJpdj0SCz3pKsFRcA entitled «General Technical Information – Film Capacitors» in which an estimation of temp rise based upon power dissipation, surface area of the capacitor and a heat transfer coefficient, alpha. Alpha for «plastic boxes with a smooth surface» is listed as 0.96 (mW)/(ºC*cm^2). That seems to apply to all the film caps I’ve ever worked with. I use similar formulas for aluminum caps, except that the formulas have an extra step where internal temperature is calculated and then extrapolated to the temperature at the top of the cap. Vishay says that temp rise, deltaT is:
deltaT = (Power dissipated in Cr) / (alpha * Surface Area in cm^2). Again, I’m assuming that this is the core temperature, not the surface temp. I calculate 6.63cm^2 of total surface area. That comes out to around 26ºC. There could be a 11ºC difference between core temp and surface temp, but this also made me curious, so I measured the actual ESR of the resonant capacitor:
The average resistance between 1 kHz and 10 kHz, where the results are real and positive, is around 50 mohm. If I go back and substitute that value into my power dissipation I get a temp rise of only 6ºC.
My goal is not to predict temperature rise exactly – in any real design I would always make extensive, actual temp readings to be sure of any component I think will dissipate some power – but I do want a set of equations that overdesign for safety margin. Right now the jury is still out on how to refine the approach I’ve just outlined.
I´m currently investigating various types of capacitors for their use at the high ripple current inputs and/or outputs of switchers. The input to a buck is the classic example. I´ve been comparing banks of pure MLCC (multi-layer ceramic caps) to mixtures of MLCC together with standard aluminum electrolytic capacitors and mixtures of MLCCs with polymer aluminum capacitors. I assumed (and yes, I know what happens when you assume…) that the electrolytics would have fairly low SRF (self-resonant frequencies) and that they´d be pretty much useless above around 200 kHz. And that´s OK – I mostly use them to damp sub-harmonic oscillations due to power supply interaction, and those subharmonic frequencies are almost always below 100 kHz.
What surprised me was that polymer aluminum devices also appear to lose their capacitance at frequencies that rarely exceed 100 kHz. I´ve measured devices at the high end of the voltage range (25V) and the low end (2V), capacitances ranging over a 33 to 470 uF range, and I´ve measured both the can-type packages (indistinguishable from SMT aluminum electrolytics unless you look very closely and see that there’s no vents) and the rectangular SMT types (D-case, E-case, etc.) I really expected the rectangular cases to have much lower ESL and therefore higher SRFs, but not a single one has an SRF higher than 200 kHz.
This is making me question the value of polymer aluminums. Yes, the ESR is very low, so they will tolerate a lot of ripple current. But they are low voltage devices, so in general you´d find them at the outputs of DC-DCs. No modern, point-of-load DC-DC regulators run at 100 kHz. In fact, most of them are touting their 1 MHz+ operation. But when I look at the plot below of this 470 uF, 2V device, all I can think is «it wouldn´t do anything at 800 kHz. Or 400 kHz. Or even 200 kHz.»
So what are these caps good for?
In theory the Design Rule Chcecker is a great tool, but in practice it´s always been like the fable about the boy who cried «wolf!» In case anyone reads this and isn´t familiar with the story: a young shepherd on top of the hill near his village gets bored and starts yelling «wolf!» to stir up some excitement, but after several false alarms the villagers stop paying attention. Then a real wolf shows up one day and when the boy sounds the alarm, no one comes.
And that´s how DRC has behaved for me for as long as I´ve used Altium. I started with Protel 98 at CPES, the lab at Virginia Tech, so that´s 16 years of false alarms, and in my last PCB design the metaphorical «wolf» finally showed up. An un-connected net was mixed in with the hundreds of other errors I´ve come to ignore over the years, and I missed it!
This put me on a quest to really dig into my design rules, mostly lifted from Altium defaults. But there´s one that I want some help with, «Minimum solder mask sliver constraint». Here´s a picture:
The violation is that there´s an area between the soldermask around the via and the soldermask around the pad of the resistor that measures less than 0.254mm.
My main question: who cares?! I´m guessing that there´s some sort of manufacturing difficulty with small slivers of soldermask. But how small can they be before a PCB gets too be too expensive or too complicated to manufacture?
16 jun 2015
I´m not sure if I´m allowed to link to commercial websites or not, so just to be safe I´ll simply state that I was reading an LT Journal entry about the LT3790, a part that can sense it´s output current with a diff amp, and at the end of the article a gain/phase injection point and method is shown for such a part when running in constant-current mode. It´s almost identical to a suggestion made by a member of the Power Supply Design Center forum when I wrote about this previously. The suggestion was to move the reference points of the probes A and B from system ground to the negative input of the diff amp. (Assuming that the signal is being injected across a resistor placed in series with the positive input of the diff amp.) Here´s a picture:
LT suggests injecting in series with the negative and referencing to the positive, but I figured it wouldn’t make much difference. After mixing up the polarity and overdriving the diff amp I settled on a 250 mV injection at 100 Hz, scaling down to 20 mV at 1 MHz, and the result seems very realistic: 1.82 kHz bandwidth with nearly 90º of phase margin. Here it is:
Next up: check this with LTspice, which will take time and patience!
15 jun 2015
This past week I finally received the two 50 uH inductors for my CISPR16-spec LISN filter. As expected they´re thinner but a good bit longer than the 5 uH devices, and I had to rearrange the common- mode choke and the first set of Y-caps and their associated resistors. I also realized that I hadn´t added connections for the output to my DUT. (Or input from my DUT…power comes in but noise comes out, so both ports are inputs and outputs, really.)
The the 50 ohm noise output port I ended up putting footprints for everything. I figure I´ll test the impedance with the AP300 netowkr analyzer and then start adding stuff, and see if the Z curve deviates enough to fail spec. I have spots for an ESD diode, two low capacitance schottkys, L-C filters and an AC coupling cap. I didn´t try to add a limiter, though. I´m going to test a bunch of circuits with the 50 ohm setting on my oscopes and have them calculate RMS power, to see if it´s likely to damage the spectrum analyzer input.
08 jun 2015
My architect and custom furniture designer buddy got the pre-cut beechwood boards last week, and someow I managed to erase the photos he sent me of the two main side pieces. But I did save some photos of the control board:
It´s a bit hard to make out, but the opening at the front in the last picture is where a drawer with the mouse and keyboard will go. The front of the drawer is the entire width of the control board, so that you can´t tell that the drawer even exists when it´s closed. This board is especially wide, so that I can comfortably play Street Fighter without elbowing the other player. Plus, a genuine Arkanoid spinner goes in the middle with buttons on either side.
When all the holes are drilled, I´ll start wiring everything up, including my custom-designed turbofire switches for all those shooting games. Those enemies in Truxton that always ended up getting me because Carpal Tunnel syndrome set in won´t know what hit’em!
More than a year ago I wrote in to the Power Supply Design Center forum on LinkedIn, asking for advice on how to use my network analyzer to measure the control loop of a switching converter-based LED driver where the current sense is differential and not referenced to ground. I think it´s fair to say that the majority of switching LED drivers use this method, and in general I like it. But I can´t figure out where to inject and measure to check the gain and phase of the control loop. Constant current sources have some differences when it comes to gain and phase, so I think the time is well invested. Either someone from the forum or perhaps the founder, Dr. Ridley suggested placing 20 to 50 ohms in series with the positive input of the internal diff amp of the LED driver IC and then injecting and measuring there. I couldn’t find any demoboards from, IC manufacturers (at least, IC manufacturers from whom I can get free demoboards…) that had a footprint for a resistor in series with the positive diff amp input, but recently I got ahold of one that does. Previous to this I tried running simulations in LTspice and got results that didn´t make any sense (barely any change in gain or phase), but doing Bode plots in LTspice is tricky and time consuming when it´s just a regular constant voltage switcher, so I wanted to see real life results. Here´s a simplified version of the schematic:
Note: this is a «floating» or «VIN-referenced» buck-boost, current flows back to the input, not to GND.
R10 is the injection resistor, «A» and «B» are where I connected the output of the injection transformer and the probes from the AP300, and if I had gotten meaningful results, I would have used «COMP» to separate the power stage and error amp. I´m sure everyone has their own methods, but when I measure control loops I do two things to make sure I´m not screwing up the operating point of the DUT: I watch the input current and I monitor the switch node on my oscope. From observing the switch node, I knew this circuit was reasonably stable, and the input current didn´t change much, nor did the switching node´s trailing edge move all the way and overtake the next pulse´s leading edge – that´s my method of noting too much signal injection – and as the following image will show, no instabilities appear in the plot.
Unfortunately, this plot means nothing to me – it´s clearly not the true gain-phase response of the control loop. It´s very slightly more interesting than the LTspice results.
So, what went wrong? The injection point seems like a good one: low Z and point B, high Z at point A. Maybe adding a matching 20 ohms in series with the negative input to the diff amp? I also thought about trying to inject a current en parallel with R4, the sense resistor for the LED current. I did try that in LTspice, actually, but got the same, meaningless results for gain and phase. Plus, I expect that the signal to noise ratio in that configuration would be terrible, since there would only be 82 mohms between A and B.
24 may 2015
In last week´s post I didn´t explain in as much detail as I wanted to about why I´m building a complex, two-stage power supply and then making all the effort to put MOSFETs in parallel to each of the four LEDs. It is complex, and it means putting a bunch of circuitry on the small MCPCB where the LED will be mounted. I could have put four 2A sync buck LED drivers, each running from the two-cell LiON battery stack – even the green LED which claims to have a Vf of 5.0V at 25ºC would still work even if the batteries were almost drained because the UVLO is set to 6.0V. So why?
The rise and fall times of the LED current in the LEDs is the answer. I need them to be very fast. The faster the rise and fall times as LED current shifts from a given LED to its parallel FET and back again, the less of the dimming cycle is taken up with the LED current in an unknown state. (Meaning, somewhere between zero and 2A.) For the best color blending, that time should be minimized. That way, the greatest percentage of the dimming cycle has the current in a known state, including at the minimum and maximum dimming duty cycles.
Normally, anything that PWM dims below about 200 Hz will produce a strobing effect, and the human eye will see it blink, especially if the light source is moving. That might be cool for the illuminated sword, but I would keep it as an effect that can be turned on and off. Then, as you get close to 1000 Hz, several components in the power supply start to vibrate as the currents shift, making an irritating hum. It´s mainly the power inductors but also any large ceramic capacitors, especially multi-layer (MLC) capacitors. Hopefully the sound effects would mask the buzzing, but again, if you´re going to do something, do it right the first time. The only choice is then to go beyond 20 kHz. If you have pets and/or you´re an animal lover, you want to go to 30 kHz to save your dog´s and cat´s ears, too. If I used four buck LED drivers and then their logic PWM pins then the rise and fall times would be very slow, and at 30 kHz I would have barely any useable dimming duty cycle at all.
17 may 2015
This is the first real project I began after setting up PID, and it sat on the back burner for a while. Recently I´ve taken things back up. The idea is to put a real, super high powered LED into my light-sabre reproductions – I have one official Darth Vader model and an original from Ultrasabers. The tricky thing is that all four LEDs go in series, like this:
The idea is to dim each LED independently, but have only one current source. In this case, a synchronous buck regulator with almost no output capacitance and a very fast, quasi-hysteretic control loop. The trouble is the gate control for the top three FETs. I´m still searching for a gate driver that can run from 19V or 20V and pump the top FET properly even if the other three are all on. (That’s unlikely, but possible.)
The current source itself is easy – first a synchronous boost regulator develops the 19V, then the sync buck LED driver and the parallel dimming FETs run from the 19V. The input is two 18650 LiON cells – the ones in your laptop battery. Plenty of A-h in those!
The LED will be a Luminus Devices’ SBM-160, which can run at a max of 2A, and should be not just bright, by blindingly so. I expect my swords to shine even in the daytime!