So this morning was spent removing some of the matrix solder points to stop HV from arcing over. So after testing with 10V and finding I had put an SMT diode around the wrong way.. I quickly moved to 60V. It passed..
Next 175V from the RCRC filter on the transformer (no limiting).
Left shows the 22.2mA through a 5.5K load resistor. The right shows output of the RCRC filter.
I need some time to explore the operation - it's current running with 10V across the 3080 so I could reduce that.
Anyway lunch ready..
Next 175V from the RCRC filter on the transformer (no limiting).
Left shows the 22.2mA through a 5.5K load resistor. The right shows output of the RCRC filter.
I need some time to explore the operation - it's current running with 10V across the 3080 so I could reduce that.
Anyway lunch ready..
I think the SDS was having a brain barf - 186dB is way below the 8bit resolution.. at least a valid result.
Here's two more - channel 1 showing the input prior to the 3080 and channel 3 showing output across the load:
AC coupled but there's definitely a drop of noise at the low end, just not quite as much as I'd expect from the 3080.
Here's two more - channel 1 showing the input prior to the 3080 and channel 3 showing output across the load:
AC coupled but there's definitely a drop of noise at the low end, just not quite as much as I'd expect from the 3080.
So current the regulator is dropping 30V across the mosfet at -4.5Vgs measured. That's dropping the RCRC output of 175V to 141.9V. Now I'm using a 5.5K 5W load that gets toasty as it's passing 26mA when the normal tube would run at 9-12mA.
The tube datasheet shows 5.3K and previously at ~175 that gave around 9-12mA with the 1.1K and 500K around it. An option here is to change the full bridge RCRC to a voltage doubler to take the 117V (actually 122 given my mains is higher than most) then double that to 245V then use the mosfet (or resistor) to scrub voltage down to 150-180V for the tube to get 9-12mA.
The tube datasheet shows 5.3K and previously at ~175 that gave around 9-12mA with the 1.1K and 500K around it. An option here is to change the full bridge RCRC to a voltage doubler to take the 117V (actually 122 given my mains is higher than most) then double that to 245V then use the mosfet (or resistor) to scrub voltage down to 150-180V for the tube to get 9-12mA.
So I'm continuing to learn/investigate the regulator and the noise. The densest blue line is 6.8mVpp noise, down from yellow 82mV.
1. I suspect, given the above when put on 20MHz BW limiting on, that alot of the noise is (a) down to RF and (b) down to my use of scope leads.
2. First the regulator seems to add noise that is not there in the input - I suspect this is due to the mosfet being close to it's on threshold and switching on/off (or one of the diodes doing the same) is creating the thin spikes you can see outside the main bottom blue trace. The regulator has enough working space and seems to be operating - so that points to the mosfet that is running at -4.5Vgs with a 1.5-4.5Vg(th) typically 3.5Vg(th) due to the incoming noise drops below the threshold.
3. The regulator is very stable and puts out almost zero heat, even with a 30V drop across the mosfet - ideally this could be lower but given I'm already putting out 26mA when I only need 9mA. The regulator output is 141.9V which is also too low, so I'm exploring ideas around that:
a) alter the mosfet to have 20V drop which means dropping the Vgs further. The input is about 0.5Vpp ripple so this may be an option although if the noise is hitting the Vgs(th) and causing output spikes that's not going to work.
b) replace the full rectified bridge with a voltage doubler incoming will be about 245Vdc and then simply burn off the few watts to 190V.
4) noise injection points I can see on the circuit:
4a) grid stopper doesn't stop on the pin of the gate mosfet, but has about 2cm of track near the input positive terminal. You can see the gate pin (yellow) here with the ripple and what appears to be ringing (perhaps the rectifiers - I need to order resistors for snubbers!) however the lower blue trace shows the regulator output with those fine spikes that appear to be precisely 1KHz. I'm slightly suspecting that this may actually be the regulator that's then causing the ripple back through to the mosfet. I have a 1u Vishay snubber (MKP) cap between the mosfet and the VIN of the regulator. The same waveform (and ripple) is output by the mosfet into the regulator (just moved the probe to check):
4b) The next option is VSET pin:
That signal is in-between the 55K and 500R resistor at the point where the Trimmer connects. That's a load of noise (~50mVpp) to what is one of the most sensitive parts of the system. I've not put a decoupling cap on the SET pin as the cap as I forgot to order one and the 32pF I have is a SMT that's too small the contacts would arc over with 200V.
So I think that would help reduce the noise considerably. I know that plus I can add some additional guard to this pin for external noise coming in.
I'm starting to get busier over the next couple of weeks so I may not have as much time to play with this but:
a) it works at 177V 🙂
b) I will order a few bits that I want to play with and the same parts for the other regulators - including parts to drop the mosfet a little more to test and for a voltage doubler.
c) a couple more clip leads.
The 140V sounds too low for a tube but that is referenced to the +20Vdc from the LV power supply. So this is at 160V already. Will have to try it once the other regulators are build - I will probably combine the LV RCRC and regulators on the PCB as they don't have the mosfet. Ideally I'd get ~180Vdc out of the regulator so maybe voltage doubler would be good.
What is nice is the entire thing seems relatively stable.
That 1KHz signal.. is there when switched off facepalm it wasn't there yesterday. Test off first..
Last edited:
Soldering in a 1KV 0.022uF FKP1 I had lying around on the SET, switching off the Mac mini and monitor and setting 20MHz bandwidth limit whilst measuring gave this ~100dBV down:
Without the 20Mhz limit shows this:
So it seems, as you'd expect in a domestic man cave there's quite a bit of noise around.
Only issue is this scope is a 8bit ADC.. so I assume it would be using the pre-amp offset then placing the 8bit range at that point.
One last one with max-hold:
That will do. I think the next steps are to order some bits with an adjustable cap, 47K trimmer to see about lowering the mosfet voltage and possibly a variable ~10pF cap to see about optimising the value on the SET pin vs transients.
Without the 20Mhz limit shows this:
So it seems, as you'd expect in a domestic man cave there's quite a bit of noise around.
Only issue is this scope is a 8bit ADC.. so I assume it would be using the pre-amp offset then placing the 8bit range at that point.
One last one with max-hold:
That will do. I think the next steps are to order some bits with an adjustable cap, 47K trimmer to see about lowering the mosfet voltage and possibly a variable ~10pF cap to see about optimising the value on the SET pin vs transients.
So some tube measurements.
I am simulating one side of the circlotron by using the Tenma as a floating supply to replace the LV supply for this test. It is providing an adjustable voltage and the positive is connected to the negative of the regulator (the entire HV is floating) which then has it's positive attached to the top of the tube. The other PS (red glow) is the 12V heater supply:
So now I can vary the LV supply and see the current draw across the tube. The Tenma current seems to read low by a 1.5mA. Instead I've measured the tube bias voltage across the cathode 500R which is is more accurate. At 30V LV giving 172 at the top of the tube, the bias reads 4.5V which is 9mA. At 20V LV that gives 4.3Vk and thus 8.6mA.
This means I test the use of 22V to lower the SOA issue with the MJE. The 12BH7A datasheets state 9mA but I've heard giving it 12mA gives less distortion.
So the current form of regulator would work but I would prefer to have the option of running at a higher voltage and a resistor would not waste too much power (ie <0.5W) difference given the current being used.
I am simulating one side of the circlotron by using the Tenma as a floating supply to replace the LV supply for this test. It is providing an adjustable voltage and the positive is connected to the negative of the regulator (the entire HV is floating) which then has it's positive attached to the top of the tube. The other PS (red glow) is the 12V heater supply:
So now I can vary the LV supply and see the current draw across the tube. The Tenma current seems to read low by a 1.5mA. Instead I've measured the tube bias voltage across the cathode 500R which is is more accurate. At 30V LV giving 172 at the top of the tube, the bias reads 4.5V which is 9mA. At 20V LV that gives 4.3Vk and thus 8.6mA.
This means I test the use of 22V to lower the SOA issue with the MJE. The 12BH7A datasheets state 9mA but I've heard giving it 12mA gives less distortion.
So the current form of regulator would work but I would prefer to have the option of running at a higher voltage and a resistor would not waste too much power (ie <0.5W) difference given the current being used.
I was thinking that perhaps some of the noise I'm getting is actually common Mode noise. There's no filtering with the return going right back to the transformer too. Especially with the ground connection on the test being at the negative pin of the regulator which I assume would lead to some ground bounce causing a effect on the SET pin. Perhaps a small common mode filter to hit that 600KHz noise on the ground pin may be good.
However I need to also look at shielding - possibly a box with a BNC connector to test the properly..
However I need to also look at shielding - possibly a box with a BNC connector to test the properly..
So I have the circlotron channel operating with the two power supplies replacing the LV PSUs for testing my hypothesis of the BJT SOA and dropping the voltage to 20-22Vdc.
First thing - this has been running for about an hour and the current has remained steady at about 50mA. The multimeters are measuring across the 0.1R resistors - also interesting is that the Tenma has no problem reading current but the Multicomp doesn't seem to read current. Perhaps being a floating supply confuses it? Perhaps the supply doesn't measure below a certain mA? The MM across the 0.1R for the multicomp BJT shows 32-36mA. Next up to find out where the difference is coming from - perhaps that tube bias that's causing the offset.
(Yes I know the leads are black-black and red-red the colour is for the supply rather than the polarity, the 60V clips are clipped to zero/ground)
The BJTs seem happy as Larry with 22V (the voltage on the LTSpice sim).
So lets hook up the scope and see how it's getting on.
First thing - this has been running for about an hour and the current has remained steady at about 50mA. The multimeters are measuring across the 0.1R resistors - also interesting is that the Tenma has no problem reading current but the Multicomp doesn't seem to read current. Perhaps being a floating supply confuses it? Perhaps the supply doesn't measure below a certain mA? The MM across the 0.1R for the multicomp BJT shows 32-36mA. Next up to find out where the difference is coming from - perhaps that tube bias that's causing the offset.
(Yes I know the leads are black-black and red-red the colour is for the supply rather than the polarity, the 60V clips are clipped to zero/ground)
The BJTs seem happy as Larry with 22V (the voltage on the LTSpice sim).
So lets hook up the scope and see how it's getting on.
Last edited:
Well it's well behaved and ran for over 2 hours before I had to shut it down for dinner.
The video is still uploading but will be available soon..
The video is still uploading but will be available soon..
https://www.analog.com/media/en/technical-documentation/application-notes/an-1099.pdf
Just been looking at noise - resistor, diode and capacitor noise for the regulators.
As I have film caps as input and output caps, i think i should turn my attention to the diode leakages and switch out thw venerable 1N4007 and zeners for something quieter.
Just been looking at noise - resistor, diode and capacitor noise for the regulators.
As I have film caps as input and output caps, i think i should turn my attention to the diode leakages and switch out thw venerable 1N4007 and zeners for something quieter.
https://www.ti.com/lit/an/slyt660/slyt660.pdf?ts=1662061850980&ref_url=https%3A%2F%2Fwww.google.co.uk%2F
Figure 4 shows the output noise is passed into the -IN to remove noise.
The LT3080 is basically an opamp and the OUT pin is connected to the -IN.
Likewise the incoming noise should also be compared to the calm OUT pin and thus cancel noise.
Noise into the SET pin is fed into +IN so in theory as long as the OUT pinis kept stable it should be cancelled out.
So if i add decoupling to the SET pin I limit the +IN SET noise but not the IN +IN. Removing the noise from out pin then it should cancel the input noise via the opamp assuming bandwidth.
https://www.analog.com/media/en/technical-documentation/data-sheets/3080fc.pdf
Looking into noise - capacitor, zener, schottky, 1/f pink noise (we’re not chopping so let’s ignore attempting to reduce this for now). The key i think is reverse noise and capacitor leakage noise. The boltzmann noise depends on resistance and spectrum spread (look for low nV/sqrt(HZ) giving our RMS noise..
So using zeners means the noise should only occur should the zener reverse conduct and if we can keep the zener out of that zone (uA) we shouldn’t be creating large amounts of noise.
If have it correct then schottky should be used wisely as they generate wide band noise across the spectrum. I assume their gradual conduction causes this to be more of an issue than zeners.
Caps seem to have leakage noise and DC bias distortion if not specified with enough charge capacity.
Figure 4 shows the output noise is passed into the -IN to remove noise.
The LT3080 is basically an opamp and the OUT pin is connected to the -IN.
Likewise the incoming noise should also be compared to the calm OUT pin and thus cancel noise.
Noise into the SET pin is fed into +IN so in theory as long as the OUT pinis kept stable it should be cancelled out.
So if i add decoupling to the SET pin I limit the +IN SET noise but not the IN +IN. Removing the noise from out pin then it should cancel the input noise via the opamp assuming bandwidth.
https://www.analog.com/media/en/technical-documentation/data-sheets/3080fc.pdf
Looking into noise - capacitor, zener, schottky, 1/f pink noise (we’re not chopping so let’s ignore attempting to reduce this for now). The key i think is reverse noise and capacitor leakage noise. The boltzmann noise depends on resistance and spectrum spread (look for low nV/sqrt(HZ) giving our RMS noise..
So using zeners means the noise should only occur should the zener reverse conduct and if we can keep the zener out of that zone (uA) we shouldn’t be creating large amounts of noise.
If have it correct then schottky should be used wisely as they generate wide band noise across the spectrum. I assume their gradual conduction causes this to be more of an issue than zeners.
Caps seem to have leakage noise and DC bias distortion if not specified with enough charge capacity.
Last edited:
I've not forgotten or stopped this project.. I've just switched focus to make a better sound testing device than the oscilloscope's 8bit ADC. It should give a decent balanced input 24bit 192KHz with good low noise floor to allow for meaning full THD testing.
I've also been considering making a set of PCBs for the power supply regulation and for a stereo board PCB using the heatsinks I've been using the for power supply. I've calculated the amp would have 16 heatsinks in total! I've also been thinking that using voltage multipliers on the tube section could also help reduce the size of the toroidal transformers with little impact to the sound. Food for thought. I can test the idea perhaps using smaller PCB style toroids if I can get the current needed out of them.
That would mean smaller power supplies (given the stereo HPA would have 9 power supplies this seems a good thing).
I've also been considering making a set of PCBs for the power supply regulation and for a stereo board PCB using the heatsinks I've been using the for power supply. I've calculated the amp would have 16 heatsinks in total! I've also been thinking that using voltage multipliers on the tube section could also help reduce the size of the toroidal transformers with little impact to the sound. Food for thought. I can test the idea perhaps using smaller PCB style toroids if I can get the current needed out of them.
That would mean smaller power supplies (given the stereo HPA would have 9 power supplies this seems a good thing).
So I've been considering the BJT thermal runaway issue. Principally as the BJT will be in proximity to a tube. The added complication is that the BJT on both sides of the circlotron will have different temperatures which then results in a voltage gradient and therefore a DC offset.
I've considered two concepts here:
1. Locating both BJTs in a back to back configuration on the same heatsink. This reduces the variance in temperature between the two differential sides. So if one gets hotter, the other gets hotter thus minimising thermal and therefore electrical difference.
2. Locating a NTC termistor/thermal diode behind each BJT, using an aluminium spacer with a channel and thermal paste. The result is we would see non-linear temperature to resistance. However its possible to linearise NTCs over a range by either:
(a) resistance based linearisation - providing a current based approach using a parallel resistor so that at room (operating temp) they match resistance and so over a range (ie our range we're interested in for standard operation) it is linear.
(b) voltage based linearisation- using voltage divider, again this becomes linear over a larger range.
You need a decent voltage regulation for each as we're talking about variations in the 0.1V range.
I thought that one option is to use a BJT to essentially monitor and ground the base of the main BJT around the bias resistor, although it needs to filter out the noise from the NTC/diode so we don't get any additional noise injected into the base of the main BJT.
I was also considering the short circuit and other scenarios where a current sense with a opamp low passfilter would then track current, so should a short occur, it can then adjust the bias to zero.
I'm still thinking this through but this could then also balance any DC offset automatically using active solid state. Balancing the tube bias could also be done in the same way.
I've considered two concepts here:
1. Locating both BJTs in a back to back configuration on the same heatsink. This reduces the variance in temperature between the two differential sides. So if one gets hotter, the other gets hotter thus minimising thermal and therefore electrical difference.
2. Locating a NTC termistor/thermal diode behind each BJT, using an aluminium spacer with a channel and thermal paste. The result is we would see non-linear temperature to resistance. However its possible to linearise NTCs over a range by either:
(a) resistance based linearisation - providing a current based approach using a parallel resistor so that at room (operating temp) they match resistance and so over a range (ie our range we're interested in for standard operation) it is linear.
(b) voltage based linearisation- using voltage divider, again this becomes linear over a larger range.
You need a decent voltage regulation for each as we're talking about variations in the 0.1V range.
I thought that one option is to use a BJT to essentially monitor and ground the base of the main BJT around the bias resistor, although it needs to filter out the noise from the NTC/diode so we don't get any additional noise injected into the base of the main BJT.
I was also considering the short circuit and other scenarios where a current sense with a opamp low passfilter would then track current, so should a short occur, it can then adjust the bias to zero.
I'm still thinking this through but this could then also balance any DC offset automatically using active solid state. Balancing the tube bias could also be done in the same way.
Last edited: