PCjr Acting Flaky

[jason]

Partial success!

Installed 2 100 uF Al organic (low ESR – 7mOhm) and it now boots KQ1 approximately 7 out of every 10 times. So, it's in a much better place now. But still not 100%. I'll try cleaning it up a little more to see if I can get it to be more reliable, but this is definitely much better! 70% of the time is better than 0% of the time!

Here is the part I used.
https://www.mouser.com/ProductDetail/647-RR71C101MDN1

Will post when I try a few more things but looking better!

[jason]

Current output with no load. Vpp is around 3 mV which I would think would be good enough...but previous testing has shown pretty clearly that the 5V is causing all my issues. I tried several of the capacitors I have laying around and none of them make a dent in the Vpp value unfortunately.

As you can see from the plot, the main frequency is this 100 kHz oscillation, but there is also a lower frequency mode going on underneath although I would be surprised if that were an issue.

I'm rather surprised that the system seems to be so hung up on this. Maybe just due to the high frequency of it and not so much the Vrms/Vpp? The original long board had higher Vpp and Vrms than this although not in this frequency range for the most part. So, I can only suspect that that is the issue?

[jason]

Okay, short update! I got some more caps in to try out and discovered that I need to clean up the 12 V power as well.

I did one of the past tests of using the original PCjr supply to supply the 12 V and the jrATX powering the 5 V (to the floppy and the M/B). And this worked fairly reliably! Although not perfect. I did get a disk read error 1 time but retrying the read immediately afterwards worked, so it's marginal. So it appears I need to try to clean up the 12 V power as well and see if I can get the 5 V a little bit cleaner.

I'll post more details as I get further but right now I've got the following on the 5V that cleans up the signal fairly well but there is still a 2 mV sine wave (Vpp) at 300kHz.
10 uF MLCC
47 uF MLCC

This has given me the best result so far.

[DoctorOctal]

Alright, here’s a run down of my testing so far.

Interesting results!

I find it interesting that the lower uF low ESR caps perform better.

I guess that means the inductive reactance (from the ESL) is more than I expected and is dominating the overall impedance at that frequency. Using a smaller capacitor reduces the ESL and the total impedance, even though the capacitive reactance is higher.

Vpp is around 3 mV which I would think would be good enough

I would think so, too, especially given what I measured below.

I got some more caps in to try out and discovered that I need to clean up the 12 V power as well.

That sounds plausible, as it was fairly noisy in your previous screenshot. Meanwhile, 2 or 3 mV peak-to-peak on the 5 V rail should be more than good enough and is getting down into the margin of error of our measurement methods.

Kudos on your progress so far!

Also, here is what the linear power supply output looks like for comparison.

Looking at your screenshot, I was thinking, “That’s got to be induced noise. There shouldn’t be switching noise like that in a linear supply, and the peak-to-peak is higher than I would expect from a linear supply.” So I took a closer look at the PSU and realized that the long board PSU is not a linear supply at all. And on the short board, only the 12 V rail is using a linear regulator. All this time I’ve just been repeating what I’ve read elsewhere and making my own ass-umptions without taking a close look. How embarrassing!

It stands to reason that there would need to be switching to generate the negative voltage rails. On the short board, the 5 V regulator is used in switching mode, and the oscillations there are used to derive -6 V via a transformer. I don’t have the schematic for the long board but can see that both 5 V and 12 V regulator chips are switching regulators, and it looks as if the -12 V is similarly derived.

I took some time to measure and characterize the ripple and noise for both the short and long boards:

• 1x probe with pig tail ground lead, placed directly on the PSU card edge
• AC coupling
• Floppy drive plugged in for load. (I was lazy for these quick measurements and just used what I had at hand. The floppy drive doesn’t provide a big load, but it was enough to get -6 V from the short board. Perhaps I’ll re-do the measurements with a better load later.)

I used the FFT feature of my oscilloscope to get a sense of the frequency distribution of the noise. For each measurement, I started with a 25 MHz range to see where the largest peaks were, then narrowed the range to zoom in on them. I’ve included one screenshot with the 25 MHz range just to show that there’s not much to see beyond 5 MHz.

Note that the 5 V and 12 V rails on both boards have ripple of less than 20 mV peak-to-peak, while the negative rails have considerably more.

Long board

The long board regulators oscillate at about 50 kHz.

5 V





12 V



-12 V



Short board

The short board’s 5 V regulator oscillates at about 20 kHz.

5 V



12 V

The 12 V rail uses a typical 7812 linear regulator and thus has very low noise (notice the scale). The 20 kHz ripple is probably induced from the 5 V rail next to it.



-6 V

[jason]

Interesting. And thanks for the update on the power supply not being linear. I'll have to make sure I quit calling it that as well .

One thing that I noticed the other day as I was poking around is that if I probe the floppy drive at the power input connector (with the system fully powered/connected), I get much higher ripple than on the power supply directly with no load, or even with the floppy drive alone out of the system. Not entirely surprising but it is drastically noisier. When I have time I'll try to capture some more plots of the jrATX vs. the long board at the floppy power connector while in the system. Or who knows, maybe I've got the peak to peak value low enough but now EMI is becoming an issue? Still a few possibilities.

Time for some more A/B testing, wheeee!

I'll try some more things out and see if I can figure it out. I may have to do more slow testing with different capacitor configurations and see how it looks in a fully connected system rather than looking at the output directly on the edge connector.

[jason]

Time for some more A/B testing, wheeee!

Had enough time for a few quick tests. All tests run with the same jrATX configuration
1) drive A (original drive) inside case - mostly loads but tends to fail when getting to first game screen with a disk error message repeatedly
2) drive A (original drive) outside case, case closed - same as 1.
3) drive B (recently fixed drive) outside case, case closed - actually loads fully sometimes, although still has weird error every now and then
4) drive C (new to me) inside case - same behavior as 1
5) drive C (new to me) outside case, case closed - mostly loads correctly! Worked almost every time

So, this is interesting. Drive A and C are both the newer designed Qumetrak (e.g. they have a proper circuit board for the resistor between the motor speed controller pin 1 and ground rather than a resistor slapped on top of the chip like drive B has). But more interestingly is that in general it behaves better outside the case. This also has me wondering if some of the bypass capacitors on the drives themselves may be getting a bit long in the tooth and having issues in this specific circumstance with just enough loss of capacitance to no longer work well with the noisier power supply. Although the results from drive A that behaves almost exactly the same inside and outside the case is not helpful.

It may be time to go back to grounded aluminum foil around the power supply again and see if that produces similar results to putting the drive inside/outside the case.

[DoctorOctal]

This also has me wondering if some of the bypass capacitors on the drives themselves may be getting a bit long in the tooth and having issues in this specific circumstance with just enough loss of capacitance to no longer work well with the noisier power supply.

I’m not inclined to think that the capacitors on the drive would have significantly reduced capacitance. The small ceramic caps might have lost some, depending on their dielectric, but I’d expect them to be pretty stable. And the larger bypass caps are tantalums, which tend to short out but otherwise are supposed to be stable over time.

I’ve preemptively replaced the tantalums across the 12 V rail on a bunch of these drives and have kept all the removed ones in a little container. Just as a sanity check, I pulled out 20 that weren’t shorted and tested their capacitance. All but 2 were at or above (mostly above) their rating of 4.7 µF. The two that weren’t were still within tolerance: 4.62 µF and 4.55 µF.

One thing that I noticed the other day as I was poking around is that if I probe the floppy drive at the power input connector (with the system fully powered/connected), I get much higher ripple than on the power supply directly with no load, or even with the floppy drive alone out of the system. Not entirely surprising but it is drastically noisier.

How much noisier? Do you have a peak-to-peak measurement?

Now that you mention it, testing at the floppy connector in a working system is a great place to test, and that’s something I can do fairly easily. So I did. Results are below.

While the rails do get noisier when using the original power boards, they stay pretty reasonable—35 mV peak-to-peak or less. There are just some 190 mV spikes on the 12 V side when the heads move to another track.

Measurement conditions:

• 1x probe with pigtail ground lead
• AC coupling
• floppy drive, adapter card, and fan all plugged in
• the three 4.7 µF tantalums on the drive’s 12 V rail have been replaced with 10 µF MLCCs
• 64 kB card and IR receiver plugged in, but no other accessories
• system running
• case open, with floppy drive resting atop the floppy adapter and the edge of the case
• built-in disk diagnostic used to exercise the drive
• probe placed in the back holes of the molex jack on the floppy drive

Long board

12 V while idle



12 V while active



12 V while seeking



5 V while idle



5 V while active



Short board

12 V while idle



12 V while active



12 V while seeking



5 V while idle



5 V while active