Yes, I measured a previous variant with simple bootstrapping of the LTP cascodes, but did not really know how to interpret the results.
Distortion seems quite low. I would say low enough, but for some academic appeal I wanted to try new techniques to further improve distortion like the differential buffered bootstrap of the LTP cascodes. I doubt that there will be any measurable difference.
I haven't found out how to discriminate between the test signal source contribution (a DAC) and the amplifier contribution to measured THD.
Since gain of my amplifier was pretty low, I had to drive the DAC to its limits, increasing THD at its output. Should be better in future with higher gain of the amplifier.
The website by Dr. Uwe Borgmann that Levonson Mark linked features an excellent article how to greatly enhance the typical sound card measurement setup: http://diy.ucborgmann.de/index.php/en/audio-messtechnik-2/thd-with-sound-card-0-0001
I consider building a bandpass filter to reduce harmonics fed into the amplifier. Apart from the harmonics, my DAC tends to output a spray of HF noise that would be attenuated by a band pass filter, too. The challenge is just that the band pass filter also needs vanishing low THD.
I'm not good at measuring things. You probably know the German saying "Wer misst, misst Mist." Well, I measured a lot of nonsense so far...
Distortion seems quite low. I would say low enough, but for some academic appeal I wanted to try new techniques to further improve distortion like the differential buffered bootstrap of the LTP cascodes. I doubt that there will be any measurable difference.
I haven't found out how to discriminate between the test signal source contribution (a DAC) and the amplifier contribution to measured THD.
Since gain of my amplifier was pretty low, I had to drive the DAC to its limits, increasing THD at its output. Should be better in future with higher gain of the amplifier.
The website by Dr. Uwe Borgmann that Levonson Mark linked features an excellent article how to greatly enhance the typical sound card measurement setup: http://diy.ucborgmann.de/index.php/en/audio-messtechnik-2/thd-with-sound-card-0-0001
I consider building a bandpass filter to reduce harmonics fed into the amplifier. Apart from the harmonics, my DAC tends to output a spray of HF noise that would be attenuated by a band pass filter, too. The challenge is just that the band pass filter also needs vanishing low THD.
I'm not good at measuring things. You probably know the German saying "Wer misst, misst Mist." Well, I measured a lot of nonsense so far...
Good idea to simulate THD again; I haven't done this for a while.
In order to provide some context, I will attach and explain the new schematic in the next post. Imagine the front end driving a diamond buffered triple that has four pairs of power transistors as presented earlier.
THD of 1kHz near clipping (~90Vpp) into a 4 Ohm load is 0.00013%
THD of 20kHz near clipping (~90Vpp) into a 4 Ohm load is 0.0013%
Clean clipping is also one of the objectives for this project. Here is a 1kHz sine wave clipped:
I'm not a big fan of simulated THD anymore because this tells little about whether given performance is attainable in reality. A super unstable amp may show stellar THD. I had so many nice THD simulation results with ampliiers that were unstable even in simulation. You may have noticed I put lots of small inductors in my simulation setup in order to somewhat approximate a real circuit. This helps to reveal some potential stability issues sometimes.
A correlation between simulation and measurement would be nice. I will put this on my to-do list.
In order to provide some context, I will attach and explain the new schematic in the next post. Imagine the front end driving a diamond buffered triple that has four pairs of power transistors as presented earlier.
THD of 1kHz near clipping (~90Vpp) into a 4 Ohm load is 0.00013%
THD of 20kHz near clipping (~90Vpp) into a 4 Ohm load is 0.0013%
Clean clipping is also one of the objectives for this project. Here is a 1kHz sine wave clipped:
I'm not a big fan of simulated THD anymore because this tells little about whether given performance is attainable in reality. A super unstable amp may show stellar THD. I had so many nice THD simulation results with ampliiers that were unstable even in simulation. You may have noticed I put lots of small inductors in my simulation setup in order to somewhat approximate a real circuit. This helps to reveal some potential stability issues sometimes.
A correlation between simulation and measurement would be nice. I will put this on my to-do list.
So here comes the new schematic bit by bit.
First I changed the constant current sources.
There was nothing wrong with the CCS unless one either power supply fails and the amp develops unhealthy DC operating conditions. I redesigned a lot of circuitry to tolerate loss of either power supply.
This looks way more complicated than it actually is.
What you see is just a "ring of two" CCS to the left that has the sole purpose to output each half's current into the other half's reference. Thus, a very stable reference is generated across LEDs D141 and D151. Attached to those reference voltages are as many cascoded CCS as needed - one for the LTP tail and two for the bootstrapping of the LTP cascodes.
There are some decoupling options, not sure yet which to install.
First I changed the constant current sources.
There was nothing wrong with the CCS unless one either power supply fails and the amp develops unhealthy DC operating conditions. I redesigned a lot of circuitry to tolerate loss of either power supply.
This looks way more complicated than it actually is.
What you see is just a "ring of two" CCS to the left that has the sole purpose to output each half's current into the other half's reference. Thus, a very stable reference is generated across LEDs D141 and D151. Attached to those reference voltages are as many cascoded CCS as needed - one for the LTP tail and two for the bootstrapping of the LTP cascodes.
There are some decoupling options, not sure yet which to install.
Further changes are that the capacitance multipliers were substituted by R-C filters as simplification and to free up PCB real estate for other circuits.
The tail CCS of the VAS is the same.
Reference voltages for clipping are generated the same way, but refer to the opposite rail now instead of ground.
Here is the actual amplification part:
Note I added some options to toy around with like OLG reduction. I don't think I'm going to use it, but addition was a low hanging fruit.
The tail CCS of the VAS is the same.
Reference voltages for clipping are generated the same way, but refer to the opposite rail now instead of ground.
Here is the actual amplification part:
Note I added some options to toy around with like OLG reduction. I don't think I'm going to use it, but addition was a low hanging fruit.
I too noticed that in simulation the THD was lower when stability was bad. Maybe a good idea to run the amp near instability.
It is clear that you can not reproduce simulated performance in the real world, it is important to build with matched precision components to come close.*
If I simulate the amp with different ccs for example, I will choose the one that gives lowest THD.
*In your schematic you have the twin cascodes with two LEDs each, that in reality will vary in voltage drop, so left and right side may have different cascode voltages, which could be an invitation for unbalance.
Weather this is an issue or not could be found out by replacing the LEDs with batteries of slightly different voltages and see if THD detoriates, matching LEDs may be required...
I simulated what happens to THD if one LED has a different forward drop by changing the color of the LED to blue. This has zero effect. The reason is that Vce is still held constant, just at another value. The output is current.
I aim for a few Volt of Vce, just enough that the transistors operate properly.
While having a simulation running, I had a look at the AC response of LTPs with different bootstrapping.
Bootstrapping from the tail seems the safest option, but also shows some peak that goes away completely with 1k base stoppers for the cascodes.
Both differential bootstrapping options from the emitters show a tendency to generate huge gain peaks.
In both cases there are several things that help to get this under control:
Base stoppers help a lot in the buffered variant. It shows that my guess of 1k both for the buffers and the cascodes is just right.
The Hawksford style bootstrap becomes more stable with 1k base stoppers for the cascodes, too, but still shows a huge peak.
Bypassing the diodes with capacitors does help, but only a little. I first guess was that the diodes have way too low output resistance, which limits the effect of the bypass capacitor. Replacing the diodes with resistors to generate the same voltage offset has surprisingly little effect. Increasing the base stoppers of the cascodes to 10k does flatten the curve, but flattening the curve causes collateral damage worse than the peak.
The peak in the Hawksford variant is independent of the current through the offset diodes.
Replacing the LTP transistors with CFPs further degrades stability.
Seems like I need to chew on this a bit longer. Bedtime now.
I aim for a few Volt of Vce, just enough that the transistors operate properly.
While having a simulation running, I had a look at the AC response of LTPs with different bootstrapping.
Bootstrapping from the tail seems the safest option, but also shows some peak that goes away completely with 1k base stoppers for the cascodes.
Both differential bootstrapping options from the emitters show a tendency to generate huge gain peaks.
In both cases there are several things that help to get this under control:
Base stoppers help a lot in the buffered variant. It shows that my guess of 1k both for the buffers and the cascodes is just right.
The Hawksford style bootstrap becomes more stable with 1k base stoppers for the cascodes, too, but still shows a huge peak.
Bypassing the diodes with capacitors does help, but only a little. I first guess was that the diodes have way too low output resistance, which limits the effect of the bypass capacitor. Replacing the diodes with resistors to generate the same voltage offset has surprisingly little effect. Increasing the base stoppers of the cascodes to 10k does flatten the curve, but flattening the curve causes collateral damage worse than the peak.
The peak in the Hawksford variant is independent of the current through the offset diodes.
Replacing the LTP transistors with CFPs further degrades stability.
Seems like I need to chew on this a bit longer. Bedtime now.
Think youre Q9 and Q10 have to low UCE. the double current sorce could have either adjusted values ad 1K bases to psu as this would drop voltage and emitter resistors would bee redused in your "double current sorces".....Or get the bases of the sorces from V amp stage. The Ubc would be abt 1-1.2 V. And for the bc5** that ist just in the unlinear delta Ib delta Ic (Ib1- Ib0)/(Ic1-Ic0)
The R19 R20 would be abt 100 Ohm.
I think some of your values would need to bee adjusted. to make it work properly. Or?
Edit put 2* 1n4148 in series withe Q11 Q12 emitters
The R19 R20 would be abt 100 Ohm.
I think some of your values would need to bee adjusted. to make it work properly. Or?
Edit put 2* 1n4148 in series withe Q11 Q12 emitters
Last edited:
I toyed around with bootstrap options and came to some general observations:
I'll keep the footprints for the capacitors tough.
While the CFPs on their own seem to be stable, variants with CFPs cause a slight peak and may even show a huge peak in conjunction with some peaky bootstrapping options.
Thanks for the tip regarding the cap across the current mirror. This seems to work with some mirrors, but not all. The one it works best with is the beta enhanced CM. As little as 10pF across the CM null the slight peak it shows without and this method maintains good bandwidth. Bob Cordell recommends a few hundred pF from base to emitter of the helper transistor, but this significantly reduces bandwidth.
Here is what Bob Cordell writes about the differential current mirror load (ref-des fit the schematic I posted):
The key advantage of this circuit is that it allows the use of a balanced current mirror structure to load the input stage. The differential current mirror exhibits very high impedance in the differential mode, but rather low impedance in the common mode. It thus provides some additional common-mode rejection. The primary elements of the differential current mirror are current sources Q9 and Q10. Emitter followers Q11 and Q12 jointly act as the current mirror helper transistors, creating and feeding back a common mode voltage to control current sources Q9 and Q10. The emitter followers also buffer the differential signal before application to the VAS LTP.
The differential current mirror establishes a well-defined common-mode voltage level to be applied to the VAS LTP. This, combined with the differential drive of the LTP, allows the use of a simple resistor tail for Q13 and Q14.
I added a separate CCS for the VAS LTP (Q13 and Q14) because I observed that any slight variation of the bias in the first LTP has dramatic effect on the VAS bias, which has impact on the OPS bias. My idea was to just establish more reliable DC conditions. Michael Kiwanuka pointed out further benefits in his paper (discussed earlier).
The voltage drop across R19 and R20 sets the bias of Q11 and Q12 (5mA in my case). Lowering R19 and R20 reduces the bias and eventually leads to violation of the compliance voltage of the VAS CCS. With this Q9 / Q10 arrangement being a current mirror I see no way to increase the voltage across Q9 and Q10.
- Using diodes to offset the cascodes causes far worse gain peaks than resistors.
- Parallel capacitors seem to make the peak even worse regardless of the offset component (contradicting my previous observation).
I'll keep the footprints for the capacitors tough.
While the CFPs on their own seem to be stable, variants with CFPs cause a slight peak and may even show a huge peak in conjunction with some peaky bootstrapping options.
Thanks for the tip regarding the cap across the current mirror. This seems to work with some mirrors, but not all. The one it works best with is the beta enhanced CM. As little as 10pF across the CM null the slight peak it shows without and this method maintains good bandwidth. Bob Cordell recommends a few hundred pF from base to emitter of the helper transistor, but this significantly reduces bandwidth.
Here is what Bob Cordell writes about the differential current mirror load (ref-des fit the schematic I posted):
The key advantage of this circuit is that it allows the use of a balanced current mirror structure to load the input stage. The differential current mirror exhibits very high impedance in the differential mode, but rather low impedance in the common mode. It thus provides some additional common-mode rejection. The primary elements of the differential current mirror are current sources Q9 and Q10. Emitter followers Q11 and Q12 jointly act as the current mirror helper transistors, creating and feeding back a common mode voltage to control current sources Q9 and Q10. The emitter followers also buffer the differential signal before application to the VAS LTP.
The differential current mirror establishes a well-defined common-mode voltage level to be applied to the VAS LTP. This, combined with the differential drive of the LTP, allows the use of a simple resistor tail for Q13 and Q14.
I added a separate CCS for the VAS LTP (Q13 and Q14) because I observed that any slight variation of the bias in the first LTP has dramatic effect on the VAS bias, which has impact on the OPS bias. My idea was to just establish more reliable DC conditions. Michael Kiwanuka pointed out further benefits in his paper (discussed earlier).
The voltage drop across R19 and R20 sets the bias of Q11 and Q12 (5mA in my case). Lowering R19 and R20 reduces the bias and eventually leads to violation of the compliance voltage of the VAS CCS. With this Q9 / Q10 arrangement being a current mirror I see no way to increase the voltage across Q9 and Q10.
Hi, here is scetch of where i wold use as current reference piont. To improove the Uce Q1, Q2.
I really belive R7 is missing in the original ! And no R3 Or R4
At R7or at R10.
At R7 Voltage drop R5//R6 and R7 makes Uce Q1, Q2 "adjustable" adjusting R7 . Tyical R7 = R5 and R6 in paralell.
If at R10 and adj R11 to make Uce min .
R11 if ref at R10. R11makes Uce min "adjustable"
(i also find that 2sa970 or equiv has lower Uce min than bc5**).
Here is the scetch:
I really belive R7 is missing in the original ! And no R3 Or R4
At R7or at R10.
At R7 Voltage drop R5//R6 and R7 makes Uce Q1, Q2 "adjustable" adjusting R7 . Tyical R7 = R5 and R6 in paralell.
If at R10 and adj R11 to make Uce min .
R11 if ref at R10. R11makes Uce min "adjustable"
(i also find that 2sa970 or equiv has lower Uce min than bc5**).
Here is the scetch:
R Dijk, I value your contribution, but I'm having difficulties following the iteration you illustrated.
I plan to have a closer look at the differential current mirror load to better understand it and will also have a look at your suggestions then.
Currently I'm investigating a gem I found while looking for tips how to deal with Hawksford cascodes and found this:
https://www.diyaudio.com/community/...s-linearity-investigation.154670/post-1984753
Glen Kleinschmid presented a clever way to bootstrap the clamp diodes attached to the VAS and this might something for me.
So far I failed to make the boostrapped clamps outperform the simple ones, but maybe I'll figure it out.
But I found out that the resistors reverse biasing the diodes better be 100k instead of 1k (refering to the GK schematic).
I plan to have a closer look at the differential current mirror load to better understand it and will also have a look at your suggestions then.
Currently I'm investigating a gem I found while looking for tips how to deal with Hawksford cascodes and found this:
https://www.diyaudio.com/community/...s-linearity-investigation.154670/post-1984753
Glen Kleinschmid presented a clever way to bootstrap the clamp diodes attached to the VAS and this might something for me.
So far I failed to make the boostrapped clamps outperform the simple ones, but maybe I'll figure it out.
But I found out that the resistors reverse biasing the diodes better be 100k instead of 1k (refering to the GK schematic).
There was no point in being boring. I only set south measurements around the input's two current sources, the smallest VCE of the transistors.
And with such a low UCE, the capacitance C-CB C-CE will increase And you are working in an area where the linearity is terrible.
I get way too many stepping adjustments without an adjustment or two
Your results are very impressive!!
You got me thinking about this circuit and possible mutations thereof, which is not boring at all. Thanks for that. I'm toying around with the simulator and slowly start to gain some understanding.
I had a look into another textbook where the specific circuit is mentioned. Arto Kolinummi writes:
Cordell has employed a topology in which common mode feedback is included in the buffer stage. The base current of Q9 and Q10 flows through R23 and R24 and V(R19) and V(R20) define the voltage at which the buffer stage stabilizes. The circuit is very sensitive to current gain and matching of the Q9 / Q10 Vbe voltages.
I have the feeling that your suggested modifications result in totally different circuits that work totally different and also perform differently, but I did not come to any conclusions yet.
I had a look into another textbook where the specific circuit is mentioned. Arto Kolinummi writes:
Cordell has employed a topology in which common mode feedback is included in the buffer stage. The base current of Q9 and Q10 flows through R23 and R24 and V(R19) and V(R20) define the voltage at which the buffer stage stabilizes. The circuit is very sensitive to current gain and matching of the Q9 / Q10 Vbe voltages.
I have the feeling that your suggested modifications result in totally different circuits that work totally different and also perform differently, but I did not come to any conclusions yet.
R7 in the first schematic you posted seems a worthwhile addition. (Q9 / Q10 bases to VCC in my schematic).
This resistor allows to increase Vce of the transistors and else seems to maintain the properties of the original circuit by Cordell.
This resistor allows to increase Vce of the transistors and else seems to maintain the properties of the original circuit by Cordell.
Hi, the first time I saw the circuit on paper there were potentiometers for the emitter resistors in the current sources. (35-40 years ago?) It should also make matching easier.. My values are taken from the drawer and only give a hint of what I see as necessary. Also, my point is to get the transistors in the current sources to work and not throw them there. Then you could just cut them out and use two resistors. Since you are well underway, it was important that these transistors should work as intended. The happy "R7" makes it possible when the emitter resistances are set to get a compromise between Uce and not too low emitter resistance or potentiometer due to matching Ube. i belive you find your own values, as it looks like you are on "revisjon2".
For inspiration:
There is an other solution to the doubble current sorces, two resistors. The two resistor circut has a name (i dont remember), have a a look at Goldmund, two resistors replace the curret sorces.
AND
Harman Kardons solution. a current mirror with gain, one more resistor. It wll also give higher open loop gain here, but wil "destroy" som other parameters. The Harman Kardn solution is somthing worth have a half eye look at anyway when study or... ( belive the Harman input mirror froma tread here and it was used in a phono stage(????)
There is an other solution to the doubble current sorces, two resistors. The two resistor circut has a name (i dont remember), have a a look at Goldmund, two resistors replace the curret sorces.
AND
Harman Kardons solution. a current mirror with gain, one more resistor. It wll also give higher open loop gain here, but wil "destroy" som other parameters. The Harman Kardn solution is somthing worth have a half eye look at anyway when study or... ( belive the Harman input mirror froma tread here and it was used in a phono stage(????)