I wonder if the regulator output cap should be at the Sense Points rather than at the FET or the Load.
If all these coincided there would be no question to answer.
If all these coincided there would be no question to answer.
I don't know. I need to think some more about this. In the meanwhile, at the load seems to work best for hifinutnut, so let's keep this as the suggested solution. Though I wonder how much this depends on the actual individual setup.
Any experts lurking? janneman, maximus?
Last edited:
If the cap is across the MOSFET, it has some ESR and ESL relative to the sense point. If placed at the sense point, it has only its own ESR and the ESR of the sense wires.
- keantoken
- keantoken
kt, good point.
One thing to consider here is the following. Let's call the two input points of the load Vin_pos and Vin_neg. When the oscilloscope probe is connected to the Vin_pos point of the load, it shows the wave there with respect to where the ground probe is connected. The ground probe should be connected to 0 volts (ground). However, the load does not see that voltage. The load sees the difference in potential between Vin_pos and Vin_neg, which is not the same as (Vin_pos - GND).
The proper way to observe the wave seen by the load is to have an oscilloscope with differential input, so that we use the trace seen by channel A minus the trace seen by channel B.
One thing to consider here is the following. Let's call the two input points of the load Vin_pos and Vin_neg. When the oscilloscope probe is connected to the Vin_pos point of the load, it shows the wave there with respect to where the ground probe is connected. The ground probe should be connected to 0 volts (ground). However, the load does not see that voltage. The load sees the difference in potential between Vin_pos and Vin_neg, which is not the same as (Vin_pos - GND).
The proper way to observe the wave seen by the load is to have an oscilloscope with differential input, so that we use the trace seen by channel A minus the trace seen by channel B.
The probe itself has what, 22pF-47pF of parallel capacitance, and some reactance depending on the cable length? I kinda doubt this would have an effect.
What kind of load does Hifi use? Is it a MOSFET driven by square wave, wirewound inductive resistor?
Perhaps pictures would be helpful too.
- keantoken
What kind of load does Hifi use? Is it a MOSFET driven by square wave, wirewound inductive resistor?
Perhaps pictures would be helpful too.
- keantoken
I gave the manufacturer part number at the Salas v1 thread. The first item in the list and you can get the datasheet from the link: Film | Digi-Key
They have now been sold out at Digi-key.
I think a picture illustrates better:
An externally hosted image should be here but it was not working when we last tested it.
If the regulator output is more inductive than the wire inductance, a capacitor placed near the regulator output may cancel the inductance, making it more stable. Or doesn't it?
With regards to different types of capacitors, it is possible that some ESR is required for damping, as Andrew T suggested. But I can't think that is conclusive, because 4 x 2.2uF polyester in parallel (ESR/4) still has a very low ESR, and they don't cause resonance. Is it possible that the ESL of the capacitor is playing up? possible. These 2.2uF cap is very small so inductance would be lower. The worse resonance seen was with the big 400V Solen which would technically have the highest inductance of all capacitors tested.
With regards to different types of capacitors, it is possible that some ESR is required for damping, as Andrew T suggested. But I can't think that is conclusive, because 4 x 2.2uF polyester in parallel (ESR/4) still has a very low ESR, and they don't cause resonance. Is it possible that the ESL of the capacitor is playing up? possible. These 2.2uF cap is very small so inductance would be lower. The worse resonance seen was with the big 400V Solen which would technically have the highest inductance of all capacitors tested.
There is a so-called "tunnel of death" when it comes to regulators. If the output cap has too low ESR, or even capacitance is too large, the regulator can become unstable.
"tunnel of death":
http://www.diyaudio.com/forums/aksa/152031-aspen-headphone-amp-2.html#post1932418
And possibly informative:
http://www.diyaudio.com/forums/aksa/152031-aspen-headphone-amp-2.html#post1932371
- keantoken
"tunnel of death":
http://www.diyaudio.com/forums/aksa/152031-aspen-headphone-amp-2.html#post1932418
And possibly informative:
http://www.diyaudio.com/forums/aksa/152031-aspen-headphone-amp-2.html#post1932371
- keantoken
Hmmm, not as bad as I thought. I simmed a very simple differential-amp series regulator (almost identical to the one in the Pass Pearl phono pre), then converted that into a shunt version (with CCS in front off course) and compared the noise figures in LTSpice. They were pretty much identical and I am beginning to wonder, too, how good these simulations are.
Off course, the shunt regulator's line rejection was much better - due to the current source, but when I cascaded 2 series regulators rejection was again in the same ballpark. If anything the shunt reg had a small advantage at higher frequencies.
Thank you. Sorry I didn't remember that you already gave us the right part number.
So they aren't Ero/Roederstein caps as I thought from the colour, but Philips/BC ones and MKP not KP (Film & Foil).
I have some samples (1uF/63V and 0.1uF/63V), but they actually grey, not blue.
I have some good news and some bad news. The bad news is that although in simulation the C1 mod doesn't affect the output impedance, in the real circuit it does, very little, but it does. So, C1 should be connected right across R9 and R11, as in post #1 of this thread if low output impedance is desired at high frequencies; the C1 mod is beneficial for preamps where low noise is more important than low output impedance at high frequencies.
The good news. I've run a number of experiments on the real prototype, focusing on load rejection. The setup was as follows. Active load comprised of a power mosfet, 100mA DC, 37mA RMS AC at 50kHz. Voltage in 43V, voltage out 21V. Current limit 500mA. Remote sensing was used, with an output capacitor connected at the regulator output (across D-S shunt mosfet pins), not at the load. The distance to the load was approximately 10cm. Channel A probe connected to the load B+, Channel B probe connected to the load B-, both ground probes connected to the regulator 0V (return). Then I observed the A-B trace which indeed, showed the lowest RMS value. I have tried the n-channel version of v1 (just like v2 without cascode and buffer), and v2. The idea was to show experimentally that the load rejection improves with the cascode and buffer.
As a result of these experiments I can confirm the following:
a) Remote sensing must be used if there is any significant distance between the regulator output and the load
b) The output capacitor has to be connected at the regulator output, not at the load
c) The cascode and buffer improve load rejection; measured at 50kHz, the output ripple induced by the active load is lowered to at least a third of the ripple without buffer and cascode
d) Best to use a fast high gain transistor for Q2; I got much better results with 2N5087 for Q2, than with MPSH81. I also tried a BC560C for Q2, but the 2N5087 was still better.
I was afraid of stability issues if Q2 was 2N5087 instead of MPSH81, but everything was fine.
Another change, using two LEDs in series with R2 to set the CCS mosfet bias. Not important, really.
Here is revision 5c.
The good news. I've run a number of experiments on the real prototype, focusing on load rejection. The setup was as follows. Active load comprised of a power mosfet, 100mA DC, 37mA RMS AC at 50kHz. Voltage in 43V, voltage out 21V. Current limit 500mA. Remote sensing was used, with an output capacitor connected at the regulator output (across D-S shunt mosfet pins), not at the load. The distance to the load was approximately 10cm. Channel A probe connected to the load B+, Channel B probe connected to the load B-, both ground probes connected to the regulator 0V (return). Then I observed the A-B trace which indeed, showed the lowest RMS value. I have tried the n-channel version of v1 (just like v2 without cascode and buffer), and v2. The idea was to show experimentally that the load rejection improves with the cascode and buffer.
As a result of these experiments I can confirm the following:
a) Remote sensing must be used if there is any significant distance between the regulator output and the load
b) The output capacitor has to be connected at the regulator output, not at the load
c) The cascode and buffer improve load rejection; measured at 50kHz, the output ripple induced by the active load is lowered to at least a third of the ripple without buffer and cascode
d) Best to use a fast high gain transistor for Q2; I got much better results with 2N5087 for Q2, than with MPSH81. I also tried a BC560C for Q2, but the 2N5087 was still better.
I was afraid of stability issues if Q2 was 2N5087 instead of MPSH81, but everything was fine.
Another change, using two LEDs in series with R2 to set the CCS mosfet bias. Not important, really.
Here is revision 5c.
Hmm, I have seen even adverse behavior in the shunt with RC as we said a little ago. Changes the damping at the driver base. The Norton will not benefit much anyway bcs the tail resistor is already big compared to the cap's impedance.
I saw you cut some Johnson by relegating some voltage to Leds at the mini casc ccs, but then again the shunt cap there was swamping it anyway. Did you manage to see even a slight measurable difference in noise or ripple by that? Also what was the measured difference between cascoding with BC560 or 2N5087? I was wondering why you used MPSH81 of hfe 60 for top initially but I thought you maybe had some practical reason like too much hfe giving some oscillation crest or something. But it was just a precaution as you explained. The more hfe there the better the psrr of the cascode. Very nice checks btw. I would be puzzled if prefiltering with gyrator, cmc etc or designining for best cascode rejection at the ccs would be the most elegant way and if they would compare in dB. I take use both and max out as an answer too.😀
Edit: Also did you measure the ripple rejection difference of the base V1 version at 20kHz and latest V2? What a simple crc prefiltering could do for bringing them near at 50kHZ? That will be a hint for our gyrator itches.🙂
I saw you cut some Johnson by relegating some voltage to Leds at the mini casc ccs, but then again the shunt cap there was swamping it anyway. Did you manage to see even a slight measurable difference in noise or ripple by that? Also what was the measured difference between cascoding with BC560 or 2N5087? I was wondering why you used MPSH81 of hfe 60 for top initially but I thought you maybe had some practical reason like too much hfe giving some oscillation crest or something. But it was just a precaution as you explained. The more hfe there the better the psrr of the cascode. Very nice checks btw. I would be puzzled if prefiltering with gyrator, cmc etc or designining for best cascode rejection at the ccs would be the most elegant way and if they would compare in dB. I take use both and max out as an answer too.😀
Edit: Also did you measure the ripple rejection difference of the base V1 version at 20kHz and latest V2? What a simple crc prefiltering could do for bringing them near at 50kHZ? That will be a hint for our gyrator itches.🙂
Yes they are the right ones. I have a few 1uF/63V as well and they are also grey. The 1.1uF are blue. For the 0.1uF/63V, my first batch 5 years ago was blue, second batch 2 years ago was grey. Note they have the 5% or 2% series.
That is great. Thanks, Ikoflexer.
It was very good timing as well. I was about to hit the button for purchasing the parts from Farnell. Now I can order all the new parts in one purchase.
Did you use the 4uF MKT cap at the output when you did your last experiments? Were their legs right on the D, S of the MOSFET?
The noise is below what I can measure, so I could see no difference in reality with/without the LEDs, with C1 on the base of Q3 or across R9 and R11. My best resolution is 100uV/div and when I don't use a sine load all I see is a straight line as shown in the pictures from the previous post.
BTW, the ripple going into the regulator is quite large, but nothing shows at the output with a passive load. When the load is a sine wave, of course, it modulates the output of the regulator. That modulation is much larger than the intrinsic noise of the regulator or the ripple that comes through from the CRC filter.
My conclusion here is that those people who were stressing themselves about nanovolt noise in the regulator... well, even a 5mA sine wave in the load will wiggle the output voltage a few microvolts. It's the output impedance that is eating them alive. Just my opinion, of course.
The reason why I used the MPSH81 for both transistors was that in simulation it seemed not to matter if the hfe was lower. In practice it does, a lot actually. 🙂 For those people reading this, that like simulating a lot, unless you build the real circuit, you'll never know the truth. I say nothing new here, just would like to stress it.
BC560C vs 2N5087. To tell you the truth I expected better results from the BC560C. I mean, better NF, better fT, higher gain. Still, the output ripple was about less than half as compared to using the BC560C.
Here's something weird. When using the 2N5087 on top, if I grab the plastic MPSH81 between my fingers, the ripple becomes significantly smaller. Why? This at 50kHz sine wave load.
Remember, this was not properly cased.
I don't understand salas, do you mean that a gyrator prefilter would be even better? I personally suspect that psrr will increase, but I already cannot measure the psrr anymore, noise is too low. I need to be able to measure below 1 microvolt. I think I will build a better noise amp and then I could probably see if gyrator prefilter would indeed help.
BTW, the ripple going into the regulator is quite large, but nothing shows at the output with a passive load. When the load is a sine wave, of course, it modulates the output of the regulator. That modulation is much larger than the intrinsic noise of the regulator or the ripple that comes through from the CRC filter.
My conclusion here is that those people who were stressing themselves about nanovolt noise in the regulator... well, even a 5mA sine wave in the load will wiggle the output voltage a few microvolts. It's the output impedance that is eating them alive. Just my opinion, of course.
The reason why I used the MPSH81 for both transistors was that in simulation it seemed not to matter if the hfe was lower. In practice it does, a lot actually. 🙂 For those people reading this, that like simulating a lot, unless you build the real circuit, you'll never know the truth. I say nothing new here, just would like to stress it.
BC560C vs 2N5087. To tell you the truth I expected better results from the BC560C. I mean, better NF, better fT, higher gain. Still, the output ripple was about less than half as compared to using the BC560C.
Here's something weird. When using the 2N5087 on top, if I grab the plastic MPSH81 between my fingers, the ripple becomes significantly smaller. Why? This at 50kHz sine wave load.
Remember, this was not properly cased.
I don't understand salas, do you mean that a gyrator prefilter would be even better? I personally suspect that psrr will increase, but I already cannot measure the psrr anymore, noise is too low. I need to be able to measure below 1 microvolt. I think I will build a better noise amp and then I could probably see if gyrator prefilter would indeed help.
Here is revision 5c.
Hi ikoflexer!
In this revision 5C Can you use for hight voltage??
Ex: 250v 50MA? for PRe!
thank!
Hi ikoflexer!
In this revision 5C Can you use for hight voltage??
Ex: 250v 50MA? for PRe!
thank!
Grab with fingers...hmm effectively sinking it? Try a TO-92 sink on it, may it will show better ripple?
I am afraid of the output impedance of the gyrator creating more hum pick up when filtering to the ccs too. That is why I wonder, will it cut high, but increase mains related harmonic noise down noise floor bottom? What about just CLC prefilter? Do you still have them on test? Can you slap a coil and see what happens? Also I did not understand, the 560C gave worse ripple by 6dB than when using the 2N5087 for top tier cascode? If yes, maybe its down to product quality of the makers because I don't see big differences in their datasheets where it counts for that. Or it has less Rbb?
I am afraid of the output impedance of the gyrator creating more hum pick up when filtering to the ccs too. That is why I wonder, will it cut high, but increase mains related harmonic noise down noise floor bottom? What about just CLC prefilter? Do you still have them on test? Can you slap a coil and see what happens? Also I did not understand, the 560C gave worse ripple by 6dB than when using the 2N5087 for top tier cascode? If yes, maybe its down to product quality of the makers because I don't see big differences in their datasheets where it counts for that. Or it has less Rbb?
About 10mm away from the D, S pins. But the cap has long leads, about 6cm. I was not particularly careful with this prototype, so, good results even so sloppy. Will redo it with a proper pcb, short leads, etc., when the time comes.
The same topology yes, but not exactly. I have a high voltage version in the works, but don't hold your breath.
- Status
- Not open for further replies.