This is post 8 in the series (previous: Fan Controller Design).

This post is sponsored by PCBWay.
Check and/or participate in their PCBWay 7th Project Design Contest.

---

With all design decisions done in the previous post, it was time to get PCBs made. As PCBWay was interested in the project, it was an easy choice. :) That said, if you go to my GitHub, you'll notice I have many more electronics projects that I haven't gotten around to write about and they are not sponsored by anybody (patience of my wife, excluded). Yet still, I quite often go for PCBWay anyhow.

In this particular case, PCB was simple enough that any manufacturer would be able to produce it. Or so I thought before PCBWay contacted me due to an issue. You see, my KiCAD export workflow relies on a script. Thus, usually results are quite repeatable. So I thought it was a false positive. But nope - PCBWay was right, I had some of my layers swapped - darn copy/paste when I was updating the script. I am not saying that some other manufacturers wouldn't catch this, but it wouldn't be a surprise if I got wrong PCBs due to this mistake on my part. With PCBWay, I never managed to sneak an error by them.

Once the board arrived, I started creating firmware and the first step was generating the 25 kHz PWM signal. A few initialization settings later, I had a perfect 50% duty cycle signal. So I connected my fans only to be surprised by their noise. I mean, I expected some noise - those are server fans after all. But I also expected that noise at 50% wouldn't be that bad. Then again, they did move a lot of air, so I decided to bring PWM down to 10%. And... fans were equally loud and moved an equal amount of air.

It took me a few visits to the fan datasheet to notice one discrepancy. Unlike normal PC fans that have PWM on wire 4, Delta's have PWM on wire 3. Wire 4 is the tachometer output. Why the heck they opted for a non-standard pinout is beyond me; but I am sure there is some historical reason behind it. In an case, I swapped wires to be in a standard fan arrangement and what I got was almost silent cooling.

Since I don't know how much cooling I'll need, I opted to have 3 fans together. And while this was probably an overkill because these beasts move air like there's no tomorrow, it also enabled me to keep fans at low speed by default, going up in speed (and noise) only when temperature requires that.

So, after consulting the power supply datasheet, I decided on the following curve:

• 10% speed until it reaches 40°C
• speed linearly increases up to 50% between 40-60°C
• stay at 50% up to 70°C
• if above 70°C, go to full 100%

The idea behind this is to balance fan noise and cooling, unless we get into temperatures that might jeopardize power supplies. Since my selected power supplies shutdown at 85°C, giving them some cooling buffer seemed like a good idea. Mind you, I don't expect them to ever reach that temperature, but it's nice to know that fans will do their best, if they do.

Those reading fan datasheet will notice that, while fans work just fine at low RPM, they do require 30% PWM to start. Such requirements are the reason why, on servers, you will hear fans running at 100% for a few seconds before slowing down to normal speeds - they have to ensure successful startup. For me, I didn't need to be that aggressive but I still wanted to give them a decent chance. So, every time power is applied, the fan controller will push 30% PWM out for 500ms, followed by a 250ms burst to full speed. After that, speed goes down to 10% and temperature control loop takes over. Just to be fancy, not all fans spin at the same time, but are offset by 250ms. There is literally no reason for this other than it giving an interesting sound profile.

I mentioned temperature a few times now, so you might be wondering how I am measuring it. The nswer is the MCP9701A temperature sensor. It gives you a temperature range from -40°C all the way to 125°C with a slope of 19.53 mV/°C. While this number looks a bit crazy, it actually translates almost perfectly into 4 bits (10-bit ADC) per °C if you use a 5V reference input. Since I don't expect sub-zero temperatures, my temperature calculation is as simple as removing 82 bits from the raw reading to make the scale start at 0°C and then dividing the whole thing by 4 (those wanting to optimize would simply use shift operation here).

Now, PIC12F1501 does have an internal temperature sensor you can use. And that would work in a pinch, but it's not without its drawbacks. The first one is that it measures the die temperature, that varies not only due to external temperature, but also based on what's happening in microcontroller itself. If you drive a lot of things, your reading will be higher even though external temperature remains the same. Other, probably related issue, is its instability. It's not what you would call a precise sensor to start with, but then you'll have to deal with smoothing out data coming out of it because it can change value by couple degrees in mere seconds. It's not that you cannot filter those things out - an long-period averaging will take care of that - but it's just a pain in the butt. Thus, I went with a proper temperature sensor.

There are some things I intentionally omitted. But one that causes the most doubt is speed readback. There are two ways I could have gotten that feature for free. One is to only have 3 channels, which would allow me to use 3 pins to control PWM and 3 pins to read back the speed. Yes, that would also mean switching to the internal temperature sensor, but I could make that work. Other way of doing that would be to have just 1 PWM signal going to all 4 fans, 4 speed inputs, and that even leaves 1 pin for a temperature input. But then the question becomes "why".

Since this board is a standalone thing, there is literally no way of letting the user know that one fan is stuck. Yes, it could have enabled an even smoother startup (since I can verify it's started instead of blindly going to 30%) but, other than that, there is simply no use for it. It would only make sense if I went onto a bigger microcontroller so I can have UART or I2C connecting it to the rest of the system, and I think that would be overkill.

And no, the system as a whole will not kill itself if fans stop spinning. My other part is controlling power supplies and that board will have its temperature sensor. If temperature goes dangerously high (e.g. because fans are not spinning), that board is in a situation to either stop overloaded power supply, inform the user, or do whatever I deem necessary in that situation. After giving it a lot of thought, I decided to keep it simple.

Source for PCBs and firmware can be found on project's GitHub pages. And in video below you can see the fan controller startup in action.

This is post 7 in the series (previous: Cooling, next: Fan Controller).

This post is sponsored by PCBWay.

---

I haven't tried to build myself a fan controller, I swear. My original plan was to get one of many PC fan boards and call it a day. However, I was surprised at how little of those fan controllers could properly work without PC. Even worse, most of their fan profiles would cause my 1U fans to scream. So, making my own became easier solution than figuring how to make the existing ones work.

First, about the fans. I had 1U case that was made of quite thick plastic. That left only 36.576 millimeters to work with. If we round down this overly precise number in order to account for some tolerances, the biggest fan I can place here is 36x36 mm in size. And that's definitely not a standard size you'll find in your desktop computer. However, it's standard enough for there to be multiple fans of that size. I opted to got with Delta FFB03612EHN primarily due to its 36x36x28 mm dimensions that would just fit into my case. But my second concern was not less valuable - it was a PWM controlled fan.

When it comes to fans, you have 3-pin ones that always run at the same speed and the only way to slow them down is to lower the voltage. This wasn't what I was interested in. I wanted 4-pin fans that have PWM signal for speed control. And my Deltas were just such fan.

To control pretty much any fan speed, you need to have a 25kHz PWM signal. There is quite a bit of latitude allowed (for Delta fans, it's usually 20-30 kHz) but different fans have different tolerances, so keeping it as close to 25 kHz is probably the best idea. Fan speed itself is not controlled by PWM frequency but by duty cycle. If the duty cycle is 50%, you run fan at 50%; if the duty cycle is 25%, fan runs at 25%; if the duty cycle is 75%, you get the drill...

Officially, a duty cycle control needs to be done by an open-drain output. That is, you only ever pull the PWM signal down, letting it go to the fan's internal 5V pull-up for the rest of a PWM cycle.

Keeping this in mind, I went to search for a microcontroller I could use for the project. It had to be low-pin count, so I don't waste a lot of PCB space; it had to have hardware PWM, so I don't need to bit-bang; it had to have open-drain PWM output; it had to have an analog input for temperature; and lastly, I had to have one available in my drawers.

Looking at Microchip PICs I had available, I quite quickly went toward the PIC12F1501. This one is a little gem. It has not one, but four PWM channels thus allowing me to control 4 fans in a completely independent manner. It also has 4 ADC channels (some pins overlap) thus allowing for an external temperature sensor (I had MCP9701A lying around). It even has an internal temperature sensor which is the fact I noticed only once I already had my PCBs made, so I ended up not using it.

But you will notice it didn't have one thing I needed - an open-drain PWM output. However, since the pull-up value within the fan is specified to be a maximum of 5.25 V, I was reasonably sure I could ignore that. Any voltage differential would be small and thus any extra current going through pull-up resistor would be way lower than the current it already had to carry during its "off" cycle when it gets connected directly to GND. In short, I was willing to ignore this part of the specification.

After messing around in KiCAD for a while, I had my design finished and it was off to my sponsor, PCBWay, to manufacture the PCBs. But let's continue that part of fan controller story in the next post.

This is post 6 in the series (previous: Power Supply Selection, next: Fan Controller Design).

---

While power supplies are selected, a huge elephant in the room remained - how are we going to cool these? Well, good news - strictly speaking, we don't have to, thanks to the miracle of derating.

If we check the datasheet, there is a derating curve where it tells us that (without any heatsinking, at 70°C, using puny 110V) we can count on about 40% of the power. So, if we don't go over 80W, everything is awesome. And I would argue that even running two NUC machines on my 15V power supply (most loaded one) will be under that.

But we can do better. The 1U case doesn't allow for a lot of natural air circulation. Even though UHP series is fine with being restricted, for their longevity, it's better if we get rid of at least some of those darn Celsius.

First thing we can add here is an aluminium plate. Adding a small 200x175x3 mm chunk of aluminium will not only do wonders for distributing heat, but also make their mounting easier. And it's surprisingly cheap these days to get a pre-cut plate with holes at the perfect location for the power supplies. I got mine at SendCutSend for $35, delivered. It's not as cheap as if you have tools and/or patience to do it yourself, but it's not going to break the bank either.

With the plate in hand, we can use M3x6 screws and nuts to fasten power supplies. Adding a bit of thermal paste and screwing 4 screws per power supply is all it takes. Yes, we're not using a plate as big as recommended in the datasheet, but this isn't nothing to laugh about. Even better, I can actually use this plate to stiffen the whole case a bit after I mount it. How? 3D printed rails sound as a good idea.

In addition, I have decided to add a few fans to keep the air circulating so we don't get even close to overheating. Realistically, as long as we have some air movement, these power supplies are going to be fine. On other hand, since I am overbuilding this thing to start with, I can build temperature-sensitive fan controller too.

But that's a story for the next blog post.

This is post 5 in the series (next: Cooling, previous: Dell Trigger).

---

When it comes to power supplies for embedded electronics, it's really hard to beat Meanwell. They are easily available, they have a wide selection, and they are reasonably priced. Thus, my search for power supply started with the first Meanwell catalog I found.

After testing the Dell trigger board, my power supply setup crystallized to 48/55V (for my routers), 15/20V for computers, and lastly (optionally) 12V for a modem. Why do I say "optionally"? Well, the option of using a buck regulator for the modem is always open. Since I have my case dimensions predetermined, the choice on whether to use 2 or 3 power supplies will be mostly driven by their dimensions.

As 1U severely restricts the height, my choice fell onto four power supply families:

• LRS-100-XX
• RSP-150-XX
• RSP-320-XX
• UHP-200-XX

I wanted to get as small as possible (can I fit 3?), a reasonable amount of power at each voltage (100W+, especially for 15V needed by computers), a reasonably small ripple (less than 200mA, if possible), tight voltage tolerance (1%, ideally), high efficiency (90%+ desired), and lastly robust overload controls (ideally with auto-recovery). Looking at the catalog, I placed the following power supplies on the short list:

Model	Output	Power	Tol	Ripple	Eff	Dimensions	Input	C Overload	V Overload	PFC	Fan	Cost
LRS-35-12	12 V 3.0 A	35 W	±1%	120 mV	86%	99 x 82 x 30	85-264 Vac	Y (auto)	Y (repower)	N	N	$13
LRS-50-12	12 V 4.2 A	50 W	±1%	120 mV	86%	99 x 82 x 30	85-264 Vac	Y (auto)	Y (repower)	N	N	$14
LRS-100-15	15 V 7.0 A	105 W	±1%	120 mV	90%	129 x 97 x 30	85-264 Vac	Y (auto)	Y (repower)	N	N	$17
LRS-100-48	48 V 2.3 A	100 W	±1%	200 mV	91%	129 x 97 x 30	85-264 Vac	Y (auto)	Y (repower)	N	N	$19
LRS-150-24	24 V 6.5 A	150 W	±1%	200 mV	89%	159 x 97 x 30	85-264 Vac	Y (auto)	Y (repower)	N	N	$19
LRS-150-48	48 V 3.3 A	150 W	±1%	200 mV	90%	159 x 97 x 30	85-264 Vac	Y (auto)	Y (repower)	N	N	$25
RS-50-12	12 V 4.2 A	50 W	±1%	120 mV	84%	99 x 97 x 36	88-264 Vac	Y (auto)	Y (auto)	N	N	$19
RSP-150-24	24 V 6.3 A	150 W	±1%	150 mV	89%	199 x 99 x 30	85-264 Vac	Y (auto)	Y (repower)	Y	N	$39
RSP-150-48	48 V 3.2 A	150 W	±1%	250 mV	90%	199 x 99 x 30	85-264 Vac	Y (auto)	Y (repower)	Y	N	$43
RSP-320-48	48 V 6.7 A	320 W	±1%	240 mV	90%	215 x 115 x 30	88-264 Vac	Y (auto)	Y (repower)	Y	Y	$54
UHP-200-12	12 V 16.7 A	200 W	±1%	240 mV	93%	194 x 55 x 26	90-264 Vac	Y (auto)	Y (repower)	Y	N	$58
UHP-200-15	15 V 13.4 A	200 W	±1%	240 mV	94%	194 x 55 x 26	90-264 Vac	Y (auto)	Y (repower)	Y	N	$54
UHP-200-24	24 V 8.4 A	200 W	±1%	240 mV	94%	194 x 55 x 26	90-264 Vac	Y (auto)	Y (repower)	Y	N	$57
UHP-200-48	48 V 4.2 A	200 W	±1%	240 mV	94%	194 x 55 x 26	90-264 Vac	Y (auto)	Y (repower)	Y	N	$59

After tinkering with a couple of combinations, including using power supplies from different families, I decided on a homogenous UHP-200-XX setup. They have ridiculously low profile, high power, and high efficiency. On the downside, they do have a bit of a ripple and their cost is quite a hit.

Another setup I could go with would still have UHP-200-15, but combined with LRS-50-12 and LRS-100-48. The downside of this approach would be a bit of a crowded central setup and less robust power supply.

The three power supplies I ended up with are UHP-200-12, UHP-200-15, and UHP-200-55. While the last one might be a surprise since I was leaning more toward 48V, the 55V version has unusually wide adjustment range (45-58V). This means it covers both 48V and 55V thus allowing me some flexibility.

If you look into their datasheet carefully, one will notice that these power supplies require quite a big heatsink. But alas, we cannot solve everything today; there needs to be something for a future me to deal with too. For now, I need to order these three before I change my mind again.