Not so shore those mos are possible to get exept fake.
But cascode for Cascode... The cap increase on p ch and decrease on N ch and oppostit. After al the unlinear Cap gate to drain chould sum kind of linnear. Or fairly konstant capasive for vas stage load.
Dont use Cascode al ower. It has smal delays and its own distortions. only one stage in cascode for eac amp.
Edit: also kaskoding here increase the gate drain cap. Cascode here you probably also need cap base output or driver.
Sorry my typing... kaskode vvs cascode and so on.
Last edited:
I just like to briefly mention how nice the dual transistors in SMD package fit in between the THT components:
In most cases it is easy to fan-out the pin-out of the SMD transistors to the THT ones quite nicely given the flat sides of the THT parts face each other.
The SOT 363 package is super tiny, which is great because it consumes very little space on the PCB, but the pitch of 0.65mm is difficult to solder.
Last time I had to do with such tiny SMD packages was over a decade ago with much better soldering equipment available and better eyes.
In most cases it is easy to fan-out the pin-out of the SMD transistors to the THT ones quite nicely given the flat sides of the THT parts face each other.
The SOT 363 package is super tiny, which is great because it consumes very little space on the PCB, but the pitch of 0.65mm is difficult to solder.
Last time I had to do with such tiny SMD packages was over a decade ago with much better soldering equipment available and better eyes.
PCB design of the new front end is done, but I need to finish two other modules prior to sending the data to the fab: The capacitance multipliers (easy) and a new revision of the simple folded drivers that form a Diamond OPS in the end. The revised driver module will be very versatile. It supports BJT or MOSFET drivers in different packages with different cooling options, Vbe or Vgs multipliers as option and works with my BJT or MOSFET OPS.
Inspired by your lateral MOSFET driver idea, I simulated a MOSFET Diamond OPS with bootstrapped drivers. This seems a very interesting option indeed. This kind of OPS draws almost no current from the VAS - compared to a Diamond buffered triple with same MOSFET OPS attached. Also, the MOSFET Diamond OPS is very simple with low part count, does not even need a Vgs multiplier.
Here is an illustration (image is a bit small, but you get the idea):
Legend: Red is the signal swing, green is the current drawn from the VAS by the DBT and blue is the current drawn by the MOSFET Diamond OPS.
Inspired by your lateral MOSFET driver idea, I simulated a MOSFET Diamond OPS with bootstrapped drivers. This seems a very interesting option indeed. This kind of OPS draws almost no current from the VAS - compared to a Diamond buffered triple with same MOSFET OPS attached. Also, the MOSFET Diamond OPS is very simple with low part count, does not even need a Vgs multiplier.
Here is an illustration (image is a bit small, but you get the idea):
Legend: Red is the signal swing, green is the current drawn from the VAS by the DBT and blue is the current drawn by the MOSFET Diamond OPS.
Hi, your dimond driver for output is with it's current sorces abt the complexity of the Cordel utput with error correction.
Not so lon ago som nice inputs for cordell error correction output stage and links...
https://www.diyaudio.com/community/threads/bob-cordell-interview-error-correction.89023/page-205
Not so lon ago som nice inputs for cordell error correction output stage and links...
https://www.diyaudio.com/community/threads/bob-cordell-interview-error-correction.89023/page-205
Yes, the part count is roughly the same.
The Cordell error correction is pretty complex and tricky to get right, but those who got it working well achieved stellar results (including Cordell).
My favorite is LKA's implementation:
https://www.diyaudio.com/community/threads/hec-amp.286565/post-6267367
Basically the HEC is a quad with parts of it wrapped inside a feedback loop. I found the triple very challenging to stabilize. The idea of building a quad terrifies me and thus like to see how much worse a dual could be.
The MOSFET diamond buffer may be a brain fart though. Built with IRF610 / IRF9610 as drivers, this presents over 300pF of capacitance to the VAS, which is rather unfortunate - bootstrapped or not.
With the redesigned module I could build nine variants to experiment with, although only some of them would make much sense:
The Cordell error correction is pretty complex and tricky to get right, but those who got it working well achieved stellar results (including Cordell).
My favorite is LKA's implementation:
https://www.diyaudio.com/community/threads/hec-amp.286565/post-6267367
Basically the HEC is a quad with parts of it wrapped inside a feedback loop. I found the triple very challenging to stabilize. The idea of building a quad terrifies me and thus like to see how much worse a dual could be.
The MOSFET diamond buffer may be a brain fart though. Built with IRF610 / IRF9610 as drivers, this presents over 300pF of capacitance to the VAS, which is rather unfortunate - bootstrapped or not.
With the redesigned module I could build nine variants to experiment with, although only some of them would make much sense:
Hi, have you had a look at this for stabilizing tempco?
https://www.diyaudio.com/community/threads/giovanni-stochino-power-amp.100556/
Edit
https://www.angelfire.com/sd/paulkemble/sound8c.html
https://www.diyaudio.com/community/threads/giovanni-stochino-power-amp.100556/
Edit
https://www.angelfire.com/sd/paulkemble/sound8c.html
Last edited:
Thanks for the link.
The Stocchino amp uses four diodes plus a precision reference in series. I could not find any information how the temperature sensing is supposed to be arranged physically, i.e. which diode senses which transistor. I would guess that two diodes are for the BJT drivers and two for the MOSFET OPS, but then the temperature coefficient does not match.
What is wrong with a Vgs multiplier for a MOSFET OPS?
What I did a decade ago was BJT drivers on their own radiators left uncompensated and use a Vgs multiplier for the MOSFET OPS. Back then I found bias pretty stable. The amp is still running and I should measure bias vs temperature again to confirm.
Bob Cordell wrote some good advice about MOSFET OPS bias stabilization in his book.
All arrangements with BJT drivers and BJT OPS suffer from low current gain. Best case this has a current gain of 10000 roughly, whereas the DBT easily has over 1000000.
The rightmost variant in the top row has rather high input capacitance just like the three all-MOSFET variants.
Probably best are the two variants with BJT drivers and MOSFET OPS.
I would need to do some simulation in order to find the best solution, but the PCBs are pretty flexible so I can procure them already.
Here is the updated schematic of the driver module:
Besides adding more options to connect the collectors of the drivers and adding a Vbe / Vgs multiplier, I also updated the CCS.
The new CCS features slightly lower compliance voltage, better SOA and is a bit simpler overall.
Note the "boot" connections would be to the OPS emitters / sources like in the lower right or second from the left schematic examples.
I guess I will keep the DBT with BJT output and pair the folded drivers with the MOSFET output, but I may change my mind and swap modules any day.
The Stocchino amp uses four diodes plus a precision reference in series. I could not find any information how the temperature sensing is supposed to be arranged physically, i.e. which diode senses which transistor. I would guess that two diodes are for the BJT drivers and two for the MOSFET OPS, but then the temperature coefficient does not match.
What is wrong with a Vgs multiplier for a MOSFET OPS?
What I did a decade ago was BJT drivers on their own radiators left uncompensated and use a Vgs multiplier for the MOSFET OPS. Back then I found bias pretty stable. The amp is still running and I should measure bias vs temperature again to confirm.
Bob Cordell wrote some good advice about MOSFET OPS bias stabilization in his book.
All arrangements with BJT drivers and BJT OPS suffer from low current gain. Best case this has a current gain of 10000 roughly, whereas the DBT easily has over 1000000.
The rightmost variant in the top row has rather high input capacitance just like the three all-MOSFET variants.
Probably best are the two variants with BJT drivers and MOSFET OPS.
I would need to do some simulation in order to find the best solution, but the PCBs are pretty flexible so I can procure them already.
Here is the updated schematic of the driver module:
Besides adding more options to connect the collectors of the drivers and adding a Vbe / Vgs multiplier, I also updated the CCS.
The new CCS features slightly lower compliance voltage, better SOA and is a bit simpler overall.
Note the "boot" connections would be to the OPS emitters / sources like in the lower right or second from the left schematic examples.
I guess I will keep the DBT with BJT output and pair the folded drivers with the MOSFET output, but I may change my mind and swap modules any day.
Have a look at the IRF640 and 9640 and baias temp.
The pcb for stochino has four sensorson the cooling fin. Just placed between outputs.
With too smal fins thatone made a termal runnaway with to smal cooling fins and a transistor istead of the pressisjon divice. (i have tried, 8 cm high cooling in a 2U monacor cabinet).
https://sites.google.com/site/steeve101uk/
The pcb for stochino has four sensorson the cooling fin. Just placed between outputs.
With too smal fins thatone made a termal runnaway with to smal cooling fins and a transistor istead of the pressisjon divice. (i have tried, 8 cm high cooling in a 2U monacor cabinet).
https://sites.google.com/site/steeve101uk/
Thanks for the link to Steve's website!
Now it is much more clear to me.
I first didn't notice that the drivers are small signal transistors in TO-92 package. The temperature of the drivers also affect OPS bias and while their bias is not compensated in any way, they are close to the source resistors of the OPS, which is not a good idea in my opinion. The higher the resistors temperature, the higher the bias...
Another potential issue is the rather thin L-bracket that is supposed to spread the heat. Chances are that the four diodes sensing the temperature do not really sense the OPS transistors temperature, but a lower temperature instead.
One advantage of the folded drivers over normal emitter followers is that their temperature coefficient is inverse. This is why a Diamond buffer does not need a Vbe multiplier. Of course this is also an advantage in case the drivers are on their own small radiator for mixed technology buffers like BJT drivers and MOSFET OPS.
Now it is much more clear to me.
I first didn't notice that the drivers are small signal transistors in TO-92 package. The temperature of the drivers also affect OPS bias and while their bias is not compensated in any way, they are close to the source resistors of the OPS, which is not a good idea in my opinion. The higher the resistors temperature, the higher the bias...
Another potential issue is the rather thin L-bracket that is supposed to spread the heat. Chances are that the four diodes sensing the temperature do not really sense the OPS transistors temperature, but a lower temperature instead.
One advantage of the folded drivers over normal emitter followers is that their temperature coefficient is inverse. This is why a Diamond buffer does not need a Vbe multiplier. Of course this is also an advantage in case the drivers are on their own small radiator for mixed technology buffers like BJT drivers and MOSFET OPS.
Small update from me:
I built the revision four of the front end and I'm still doing measurements. So far, it doesn't look too bad.
Due to some stupid simulation mistake, I accidentally changed the compensation to ~1MHz UGLF, which is a bit high IMO. I consider lowering this a bit.
However, here are some metrics:
At 80Vpp into 8 Ohm load:
1kHz
Miller 150p+68p THD is 0.00059%
TPC 150p 2k8 68p THD is 0.00062%
TPC 150p 2k 68p THD is 0.00055%
Not much difference here.
At 10 kHz, the picture changes:
Miller 150p+68p THD is 0.0029%
TPC 150p 2k8 68p THD is 0.0019%
TPC 150p 2k 68p THD is 0.0019%
More interesting in my opinion than some numbers, is the harmonic spectrum.
At such high levels, the Topping DAC outputs quite some harmonics and some seemingly unrelated HF trash.
I built a second order active filter to get rid of this prior to feeding the amplifier, but the filter does not work well apparently, thus I don't use it at the moment.
Instead, I just installed a cap that makes a first order filter roll off at 1.5kHz / 15kHz so that there is some attenuation at least.
Here is the spectrum near clipping:
The h2 must stem from my amp entirely since the DAC does not output any. Instead, the DAC mostly outputs h3 and I guess this is why h3 is highest. Beyond h4, the passive filter seems to do a good job filtering the input and the amp seems not to add a lot of higher order harmonics on its own. This is what I wanted to achieve.
Overall, the DAC has 0.00019% THD at 1kHz and the level used for the measurement. May I just subtract this number? Probably not.
Still difficult to say how much exactly each device contributes.
The mains voltage and harmonics are likely related to the messy setup. Sometimes it is better, sometimes worse. I bet you would call pest control in case you could see the messy cable tangle on my desk - mains, audio, some SMPS, monitor - all tangled together into a Gordian knot.
I built the revision four of the front end and I'm still doing measurements. So far, it doesn't look too bad.
Due to some stupid simulation mistake, I accidentally changed the compensation to ~1MHz UGLF, which is a bit high IMO. I consider lowering this a bit.
However, here are some metrics:
At 80Vpp into 8 Ohm load:
1kHz
Miller 150p+68p THD is 0.00059%
TPC 150p 2k8 68p THD is 0.00062%
TPC 150p 2k 68p THD is 0.00055%
Not much difference here.
At 10 kHz, the picture changes:
Miller 150p+68p THD is 0.0029%
TPC 150p 2k8 68p THD is 0.0019%
TPC 150p 2k 68p THD is 0.0019%
More interesting in my opinion than some numbers, is the harmonic spectrum.
At such high levels, the Topping DAC outputs quite some harmonics and some seemingly unrelated HF trash.
I built a second order active filter to get rid of this prior to feeding the amplifier, but the filter does not work well apparently, thus I don't use it at the moment.
Instead, I just installed a cap that makes a first order filter roll off at 1.5kHz / 15kHz so that there is some attenuation at least.
Here is the spectrum near clipping:
The h2 must stem from my amp entirely since the DAC does not output any. Instead, the DAC mostly outputs h3 and I guess this is why h3 is highest. Beyond h4, the passive filter seems to do a good job filtering the input and the amp seems not to add a lot of higher order harmonics on its own. This is what I wanted to achieve.
Overall, the DAC has 0.00019% THD at 1kHz and the level used for the measurement. May I just subtract this number? Probably not.
Still difficult to say how much exactly each device contributes.
The mains voltage and harmonics are likely related to the messy setup. Sometimes it is better, sometimes worse. I bet you would call pest control in case you could see the messy cable tangle on my desk - mains, audio, some SMPS, monitor - all tangled together into a Gordian knot.