Finally I was able to do some electronics shopping and got the few (trivial) parts that I needed. They weren’t even fancy, just some jumbo crocodile clips and a few resistors and miscellaneous parts. Took me only a month!
Anyway, as I returned to my bench, I noted my pre-regulator already set up and remembered I needed to run some tests. Last time I had worried I had used a PNP instead of NPN darlington transistor by mistake and found that I was indeed using NPN and it was working just fine. I also noted some disturbing voltage readings when heavily loading it.
In the last post, the pre-regulator worked just fine with a bit of voltage variation over the range of loads I could be using. If using a high resistive load, the voltage could climb too high above 30V, dangerously increasing the voltage differential the poor LM317 has to deal with (which could cause an overload even with a pass transistor taking most of it), or too low when using a low resistance load (higher current) dipping below my absolute minimum required of 26V.
The maximum user selectable voltage output of my power supply will be 24V (actually 48V as it is ±24V and there is a negative mirror of the positive side, we’re just dealing with one side for now) so the LM317 needs at least 2V of dropout (or headroom) to output a good clean 24V which is how I arrive at 26V. Of course, electronics being electronics, I know better than to call it a day once I get 26.0V, a bit of padding needs adding.
I had previously measured 25.something volts when using a ~3A load (my home made resistor) which is below the minimum I need. Most frustrating. I postulated a bit of capacitance could beef it up so I gave it a go and tried a bunch of caps before settling on 100µF which gave me a nice 27.4V which is just plain ideal! Problem solved there. For now. I checked with higher resistance loads (up to 10kΩ) and found it gave me an output of 30.4V which is still acceptable to me.
Of course, this is only one piece of the puzzle. I’m going to have to check again to see how well it fares with the current limiter, then voltage regulator, then load after that. It is more than likely some tweaking will be necessary to keep it all nice and stable and happy.
Of course, just getting the right voltage and current out isn’t the whole story. My DMM updates kind of slowly, and at the 200V range I have to use, it gives me only one decimal point to work with below 1V so it’s hard to see quick or small variations in voltage over time. Naturally, I fired up my ancient oscilloscope to take a peek at what it’s actually outputting. To my understanding, no power supply is perfect, and some ripple is always present. What is acceptable ripple I have no idea but I always assume the less the better. Considering a 30V output, the unfiltered ripple would be 30V which is insane. Thanks to the huge filter caps I have on the thing I can reduce that a fair deal. This is the first time I tried to measure the ripply on my power supply.
The result seems to be approximately 6mV in a curve that screams to me capacitor which is not surprising. A quick google search turns up a ripple value for your average computer PSU of 120mV on the 12V line which is twenty times what I measured from my home made power supply! Having measured my own hacked ATX PSU it was also noisy as hell, which I hear is endemic to switch-mode supplies. Sure I had my 6mV of ripple, but at least it was clean ripple! A nice clean waveform. It seems I’m on the right path. More measurements to come later as I add other modules. The later stages and in particular the voltage regulator *should* show even less ripple. I remember a section on the LM317 on ripple rejection…
I will also probably double check my results. I do remember I saw a Fundamentals Friday episode of Dave Jone’s excellent EEV blog on the very subject of measuring ripple and noise of a power supply, found here.
Another consideration was, with how hard I’m driving my transistors by pulling 3A through them, was heat. I have zero idea on how to calculate how big a heat sink I need for a particular application and naturally assume the bigger the better. I’m sure Dave Jones has something on that also come to think of it. Time to get back to class. In the meantime, because I hate leaving the bench when I can melt things straight away, I used my indefatigable logic to conclude that I will drive it as hard as I plan to run it and take a bunch of measurements. Then I will plan for at least 20% margin above that.
I used the thermocouple that came with my multimeter to measure the case temperature of the TIP142 transistor and ran it through a 3A load. The temperature climbed steadily after a few minutes to about 120°C which is bloody hot! I’m uncertain whether or not it would have climbed higher than that given time, but it was climbing very very slowly at that point. 120°C is exactly 80% of the TIP142’s max junction temperature of 150°C which sounds just fine but I’m not really satisfied. If I left the thing running for three hours would it approach or exceed 150°C? I have no idea. Better over-safe than melted and sorry.
I will definitely be using a larger heat sink for all six pass transistors used in my design, and of course there will also be forced-air cooling as well. Pretty much all semiconductors tend to start acting badly as they approach their maximum rated temperatures so it’s a good idea to keep them as cool as possible. Easily done.
If you are a budding hobbyist reading this I recommend building your own power supply, you’ll learn a hell of a lot in the process.
Have a good look at the pic of my oscilloscope reading above. Note that it is set to 2ms/division. So each of those squares on the x-axis is 2ms. Note that there are four (close enough) divisions for each wave. 4 divisions at 2ms/div is 8ms. Now think of this. The AC cycle in our houses, which I have converted to DC in my power supply, is 60Hz or oscillating 60 times per second. Because I use a full-wave bridge rectifier, that frequency doubles because it inverts the negative voltage to positive voltage so it oscillates at 120Hz. So what is the wavelength? 1 second / 120 times a second is 0.0083333ms. So it will take 8 and 1/3 ms for each wave which is exactly what you see on my ‘scope :).
Update part two
To give you a comparison, in the video linked above about power supply ripple and noise, Dave Jones compares the data sheets for two power supplies, one a high current switchmode, the other the Rigol DP832 commercial bench supply. The former had a 10mV rms ripple rating, and the latter had a 2mV peak-peak rating. If i am reading my oscilloscope correctly, I am showing 6mV peak to peak to I am actually very happy with that result. When I have more of it built, I will have to of course test it under a wide variety of loads and get a more accurate picture of the ripple coming out of it but it is encouraging.