Well, we all have different ideas and philosophys. My take is to reduce or eliminate the 2 x 5R and make sure your layout is conducive to getting the 7824 to do most of work with ripple reduction. Make sure the ground/earth/-ve wire/track is such that the 7824 takes it's feed from the capacitor bank.
Especially now YMMV Dave
Especially now YMMV Dave
Last edited:
Thanks, Dave. Not sure if what I'm up to rises to the level of philosophy just yet 🙂
Pondering your advice, it occurred to me that the longest run on my breadboard is between the filter and rectifier -- it's so long in fact that while I was thinking the options through, I kinda missed the fact that my .33 regulator input cap is effectively paralleled across a 4700uF filter cap already. I could put a bigger one in that position, but it costs me nothing to just take R2 out of the filter, see how much that affects ripple on the output (or IF it does...) and then see if the damn thing starts up.
I'll report back.
Pondering your advice, it occurred to me that the longest run on my breadboard is between the filter and rectifier -- it's so long in fact that while I was thinking the options through, I kinda missed the fact that my .33 regulator input cap is effectively paralleled across a 4700uF filter cap already. I could put a bigger one in that position, but it costs me nothing to just take R2 out of the filter, see how much that affects ripple on the output (or IF it does...) and then see if the damn thing starts up.
I'll report back.
Thanks for the kind words.
Last night: Removed R2 from the filter. Series resistance is now 5 ohms (+ESR, etc). I also drastically shortened the physical distance between the end of the filter bank and the regulator input. Drum roll, pls...
Image 1: Startup to 33rpm. Image 2: Startup to 45rpm. Image 3: Startup to 78rpm
Undervoltage issue solved. All three speeds start up like a dream, and I can change speeds at will while the platter is running. Sag after the switch throw is greatly reduced (~2v), the magnitude is the same regardless of speed selection, and I've got a clear picture of how long it takes for the controller to get a handle on things and settle into the nominal RPM.
For 33 and 45, the magnitude of the overshoot is diminished (now 6v, vs 8v overshoot before). But I expected the 78 overshoot to be equal if not greater - consistent with the DMM current measurements I took using the SMPS. But instead it's halved. Any ideas as to why that is?
This all looks like good news. With one important caveat – according to RPM app, the deck runs about .01 faster, and the W&F % has doubled. Yes, it's still low, but if its higher than it is with the switcher, all this effort is nil. Zoomed in to 20mV/div, the ripple looked... different. Not much greater in amplitude, but definitely different. Tonight I'll investigate that further.
Thanks for sticking with me, Dave and everyone.
Last night: Removed R2 from the filter. Series resistance is now 5 ohms (+ESR, etc). I also drastically shortened the physical distance between the end of the filter bank and the regulator input. Drum roll, pls...
Image 1: Startup to 33rpm. Image 2: Startup to 45rpm. Image 3: Startup to 78rpm
Undervoltage issue solved. All three speeds start up like a dream, and I can change speeds at will while the platter is running. Sag after the switch throw is greatly reduced (~2v), the magnitude is the same regardless of speed selection, and I've got a clear picture of how long it takes for the controller to get a handle on things and settle into the nominal RPM.
For 33 and 45, the magnitude of the overshoot is diminished (now 6v, vs 8v overshoot before). But I expected the 78 overshoot to be equal if not greater - consistent with the DMM current measurements I took using the SMPS. But instead it's halved. Any ideas as to why that is?
This all looks like good news. With one important caveat – according to RPM app, the deck runs about .01 faster, and the W&F % has doubled. Yes, it's still low, but if its higher than it is with the switcher, all this effort is nil. Zoomed in to 20mV/div, the ripple looked... different. Not much greater in amplitude, but definitely different. Tonight I'll investigate that further.
Thanks for sticking with me, Dave and everyone.
Btw - I've been working on a plan for the actual wiring layout in the build. Currently the plan is to have the last cap of the filter, the regulator input bypass cap, and the 7824's input pin all on the same solder tab.
@mcrushing What size capacitor do you have at the 7824 output? I was looking but it's not exactly clear...
It’s 4700uF. (And I’m considering bypassing it with a couple uF film to take care of any HF junk the big cap is missing.)
Here’s a quickie schematic I drew to keep track of things while wiring it up.
Note - I’ve now removed the second 5 ohm resistor. Also, I forgot to draw the ground point for my rail return paths - but the breadboard does actually have one 😅
Here’s a quickie schematic I drew to keep track of things while wiring it up.
Note - I’ve now removed the second 5 ohm resistor. Also, I forgot to draw the ground point for my rail return paths - but the breadboard does actually have one 😅
Last edited:
Hey, guys. So I don't happen to have a 1uF cap on hand...Amazon will be delivering a couple assortments of cheap small value films and ceramics later today. But for the moment I have that 4700uF regulator output cap bypassed with a .022uF tropical fish I happened to have in the drawer (which is probably doing not much).
Anyway, I've been scoping ripple across the output of my PSU with the motor running. Above are captures at 33, 45 and 78 rpm, zoomed to 20mv/div. Below, a quickie chart of RMS and Pk-Pk noise on the 24v rail.
| Speed | RMS Noise | Peak-to-Peak | |
|---|---|---|---|
| 33 RPM | 11.03 mV | 96 mV | |
| 45 RPM | 8.34 mV | 39.2 mV | |
| 78 RPM | 21.23 mV | 132 mV |
On their own, these data aren't all that interesting. What IS interesting is that they show a fair amount less noise on the rail at 45 rpm... and EVERY single W&F measurement I've taken with the iphone RPM app has shown a BETTER W&F FIGURE at 45 than at 33.
Yes, I know, the RPM phone app isn't all that accurate, BUT... If an iphone accelerometer can see a measurable difference in W&F from a ~2.6mV RMS reduction in power rail noise, it definitely seems worth eliminating. Gonna see what some bypass caps will do.
I still don't trust the app but that's me. I'm measurement obsessive. You are doing great work and I'm sure I, 6L6, JP and many others are watching with great interest.
That´s a far too big cap on the output. Remember.... The smoothing is by far done in the 7824 itself. The startup charge of a 4700µF is not
particularly healthy for a regulator. I would use a 78S24 (1,5A), and since you already
have 3 x 4700µF on the input, replace the 4700µF on the output with a 100-470µF. Add a 100nF on both input- and output legs.
Also consider/try a diode f.ex. 1N400x to kill evt. overshoot on the output leg. Anode on output and cathode on input of the 7824.
So I reworked my lead lengths, added the diode and got a couple assortment boxes of cheap capacitors.
I started with bypassing the 4700uF cap across the regulator output at .1uF, and because I’m still on a breadboard, it was fairly easy to swap out different values and/or stack films with smaller ceramics directly across the regulator pins. Comparing each combo on the scope, here’s what worked best:
.47uF + .022uF in parallel with the 4700uF electrolytic across the output
.33uF + .033uF + 10nF ceramic across the regular input
We were seeing 8-20mV RMS yesterday, so I think these captures are a pretty big improvement, esp. considering it isn’t on a real chassis yet. (The idea of moving from CRC to CLC has crossed my mind as well.)
Also @Dave Cawley I’ve definitely moved toward your way of thinking about the phone app tachometer….but I’ll have to get this thing in a plinth with a tonearm before I can do any real W&F work. (Speaking of… does anyone know of W&F software I could run on a mac?)
I started with bypassing the 4700uF cap across the regulator output at .1uF, and because I’m still on a breadboard, it was fairly easy to swap out different values and/or stack films with smaller ceramics directly across the regulator pins. Comparing each combo on the scope, here’s what worked best:
.47uF + .022uF in parallel with the 4700uF electrolytic across the output
.33uF + .033uF + 10nF ceramic across the regular input
We were seeing 8-20mV RMS yesterday, so I think these captures are a pretty big improvement, esp. considering it isn’t on a real chassis yet. (The idea of moving from CRC to CLC has crossed my mind as well.)
Also @Dave Cawley I’ve definitely moved toward your way of thinking about the phone app tachometer….but I’ll have to get this thing in a plinth with a tonearm before I can do any real W&F work. (Speaking of… does anyone know of W&F software I could run on a mac?)
Been quiet for a minute – Busy period at work, and I'm still waiting in line for that CNC machine to build my plinth. It'll hopefully happen in the next few weeks. Over the weekend however, I moved my Linear PSU prototype onto perfboard.
Admittedly a little solder-sloppy, but minimized lead length, loop area worked wonders on the Pk-Pk noise on the rail:
Once I got this up and running, I spent an afternoon reverse engineering the pins on the ribbon that connects the main PCB to the DD motor board:
Two takeaways from this: 1. You SP-10 guys have it easy... mostly discrete parts, service manuals with actual circuit diagrams. The 1500c/1200mk7 service manuals have block diagrams that look like they were made by someone still using Windows 95.
2. This is where I admit I'm very grateful for AI. (I've really come around on this topic.) The creative ambition for this project is all me, but ChatGPT 4.0 filled the knowledge and research gaps that made it possible for me to figure this out.
Within a few hours, I had a very good idea of what each of the 11 pins on that I/O do, and I was able to scope the FG+ waveform. (FYI, my scope came with one 10x probe and one gator-clips-to-BNC, which has been all I've needed until now. Tomorrow
Admittedly a little solder-sloppy, but minimized lead length, loop area worked wonders on the Pk-Pk noise on the rail:
| 33.3 | 32.8 mV (breadboard) | 7.2 mV (perfboard) |
| 45 | 48 mV | 19.2 mV |
| 78 | 66.4 | 24 mV |
Once I got this up and running, I spent an afternoon reverse engineering the pins on the ribbon that connects the main PCB to the DD motor board:
Two takeaways from this: 1. You SP-10 guys have it easy... mostly discrete parts, service manuals with actual circuit diagrams. The 1500c/1200mk7 service manuals have block diagrams that look like they were made by someone still using Windows 95.
2. This is where I admit I'm very grateful for AI. (I've really come around on this topic.) The creative ambition for this project is all me, but ChatGPT 4.0 filled the knowledge and research gaps that made it possible for me to figure this out.
Within a few hours, I had a very good idea of what each of the 11 pins on that I/O do, and I was able to scope the FG+ waveform. (FYI, my scope came with one 10x probe and one gator-clips-to-BNC, which has been all I've needed until now. Tomorrow
Oops... Hit send on the previous post and posted it before I finished writing. ANYWAY, continued from Post #97:
...I have two better 10x's coming via Amazon tomorrow so I can compare FG+ and FG- simultaneously.
Today, I simply compared the FG+ waveforms for both the stock SMPS and my LPS design. Here we go:
1. SMPS at 33.3 rpm - 1.83 V rms
2. SMPS at 45 rpm - 3.02 V rms
3 & 4 - 33 RPM zoomed in to 500mV/div and 1ms/div. It's hard to see without video, but the peaks of the waveform were sort of "breathing" - which is to say they fluctuated in amplitude by maybe 250 mV at a period of about 1.5 seconds. You can also see the positive peak is always within a +2v range, but the negative peak pushes offscreen low, to > -2V.
Hitting the start/stop while using the SMPS, I saw frequency modulation and phase flips for a moment before the rotation stabilized.
Moving to the Linear Power Supply (same timebase, same V/div):
33 RPM - 151 mV (!!) rms
45 - 187 mV rms
Uh... wow!
Way more stable baseline, and not only is the "Breathing" (aka the low freq amplitude modulation) gone, but the FG seems to be working less hard -- A LOT LESS -- than it was with the switching power supply. Hitting start/stop, the waveform stretched out flat and then hitting it again I watched it smoothly re-coil into a wave like a metal spring.
The AI's analysis of this, in a nutshell, is that this is the "real" FG signal...and the previous one from the SMPS test ( >10x greater) was simply being amplified by the 100mV of ripple/switching noise that we saw when I scoped that rail.
Below, the 33 rpm FG waveform zoomed WAY in. There are little transients attached to the sine, which look like noise -- except they're grouped at the peaks, and kinda regular-ish in frequency. Are they maybe not noise, but artifacts of the motherboard CPU sampling this signal, or sending PWM corrections?
Whatever they are, I think we're looking at some pretty strong evidence that a cleaner DC rail has a measurable effect on FG signal quality, and by extension – servo correction behavior.
Is it audible? Maybe? But the scope doesn't lie!
I gotta think it affects transient sharpness, W&F/pitch-drift and PRaT. @Dave Cawley @6L6 I'd love to hear your thoughts on this, and whether you've had experience comparing PSU topologies with your SP-10s.
...I have two better 10x's coming via Amazon tomorrow so I can compare FG+ and FG- simultaneously.
Today, I simply compared the FG+ waveforms for both the stock SMPS and my LPS design. Here we go:
1. SMPS at 33.3 rpm - 1.83 V rms
2. SMPS at 45 rpm - 3.02 V rms
3 & 4 - 33 RPM zoomed in to 500mV/div and 1ms/div. It's hard to see without video, but the peaks of the waveform were sort of "breathing" - which is to say they fluctuated in amplitude by maybe 250 mV at a period of about 1.5 seconds. You can also see the positive peak is always within a +2v range, but the negative peak pushes offscreen low, to > -2V.
Hitting the start/stop while using the SMPS, I saw frequency modulation and phase flips for a moment before the rotation stabilized.
Moving to the Linear Power Supply (same timebase, same V/div):
33 RPM - 151 mV (!!) rms
45 - 187 mV rms
Uh... wow!
Way more stable baseline, and not only is the "Breathing" (aka the low freq amplitude modulation) gone, but the FG seems to be working less hard -- A LOT LESS -- than it was with the switching power supply. Hitting start/stop, the waveform stretched out flat and then hitting it again I watched it smoothly re-coil into a wave like a metal spring.
The AI's analysis of this, in a nutshell, is that this is the "real" FG signal...and the previous one from the SMPS test ( >10x greater) was simply being amplified by the 100mV of ripple/switching noise that we saw when I scoped that rail.
Below, the 33 rpm FG waveform zoomed WAY in. There are little transients attached to the sine, which look like noise -- except they're grouped at the peaks, and kinda regular-ish in frequency. Are they maybe not noise, but artifacts of the motherboard CPU sampling this signal, or sending PWM corrections?
Whatever they are, I think we're looking at some pretty strong evidence that a cleaner DC rail has a measurable effect on FG signal quality, and by extension – servo correction behavior.
Is it audible? Maybe? But the scope doesn't lie!
I gotta think it affects transient sharpness, W&F/pitch-drift and PRaT. @Dave Cawley @6L6 I'd love to hear your thoughts on this, and whether you've had experience comparing PSU topologies with your SP-10s.
Last edited: