One possible reason I see for better sonics of resistor is that any CCS using
active components will have an associated capacitance parelleled to its output resistance.
When one consider this very high dynamic resistance of CCS of more advanced CCS, lets say 10Mohm,
it is easy to realize that with 10pF load capacitance (active element junctions) it will cause 15kHz cut off frequency.
This is what limits the VAS high frequency response.
This capacitance is nonlinear. Using more complex CCS with the feedback (Wilson, ANF) may lead to some more complex behavior at higher frequencies, slew rates.
Especially ANF does contain deep NFB. I simulated the ANF configuration for relatively fast voltage transitions on the output (10V/us) and in the bottom transistor the current looked pretty ugly. It had to compensate for transient current flowing in CB capacitance.
That would explain why the simple CCS are probably the best compromise. I would follow good suggestions from AKSa regarding degeneration resistor.
For VAS stages it may be better to use cascode with low capacitance transistor (perhaps high voltage video amp BF922 etc)
Cheers,
Przemek
active components will have an associated capacitance parelleled to its output resistance.
When one consider this very high dynamic resistance of CCS of more advanced CCS, lets say 10Mohm,
it is easy to realize that with 10pF load capacitance (active element junctions) it will cause 15kHz cut off frequency.
This is what limits the VAS high frequency response.
This capacitance is nonlinear. Using more complex CCS with the feedback (Wilson, ANF) may lead to some more complex behavior at higher frequencies, slew rates.
Especially ANF does contain deep NFB. I simulated the ANF configuration for relatively fast voltage transitions on the output (10V/us) and in the bottom transistor the current looked pretty ugly. It had to compensate for transient current flowing in CB capacitance.
That would explain why the simple CCS are probably the best compromise. I would follow good suggestions from AKSa regarding degeneration resistor.
For VAS stages it may be better to use cascode with low capacitance transistor (perhaps high voltage video amp BF922 etc)
Cheers,
Przemek
PB
"When one consider this very high dynamic resistance of CCS of more advanced CCS, lets say 10Mohm,
it is easy to realize that with 10pF load capacitance (active element junctions) it will cause 15kHz cut off frequency.
This is what limits the VAS high frequency response."
I suppose it works like this :
Even if there is a 15 kHz cut off, the constant current source behaviour is still there : the initial 10 MOhm dynamic resistance falls to 1 Mohm at 150 kHz, 100 kOhm at 1.5 Mhz and 10 kOhm at 15 Mhz where the amplifier may have not any forward gain.
For a long tail pair, I would add a resistor in series with the CCS whose value is such that the current across it makes a voltage drop equal to half the power supply voltage. If the current through each device of the differential pair is 3 mA, with a degenerative Re of 100 Ohm, this resistor could be a 2.2 kOhm. This is about twenty times higher than the values determining the Gm of the input stage, and it is valid at all frequencies of interest. And we still keep all the advantages of the CCS.
The often advocated suppression of the CCS in non inverting amps is a mystery to me. Its immediate effect is to increase non linearity of the input stage, which means more distorsion.
Due to the Miller effect, the output impedance of the Vas main transistor is always much smaller than the CCS. Typically, it starts to decrease at 200 Hz. I've never seen a resistor in series with the CCS of the VAS, but, experimentally, it may not be a bad idea, at the expense of sacrifying the full voltage swing.
~~~~~~~~~Forr
§§§
"When one consider this very high dynamic resistance of CCS of more advanced CCS, lets say 10Mohm,
it is easy to realize that with 10pF load capacitance (active element junctions) it will cause 15kHz cut off frequency.
This is what limits the VAS high frequency response."
I suppose it works like this :
Even if there is a 15 kHz cut off, the constant current source behaviour is still there : the initial 10 MOhm dynamic resistance falls to 1 Mohm at 150 kHz, 100 kOhm at 1.5 Mhz and 10 kOhm at 15 Mhz where the amplifier may have not any forward gain.
For a long tail pair, I would add a resistor in series with the CCS whose value is such that the current across it makes a voltage drop equal to half the power supply voltage. If the current through each device of the differential pair is 3 mA, with a degenerative Re of 100 Ohm, this resistor could be a 2.2 kOhm. This is about twenty times higher than the values determining the Gm of the input stage, and it is valid at all frequencies of interest. And we still keep all the advantages of the CCS.
The often advocated suppression of the CCS in non inverting amps is a mystery to me. Its immediate effect is to increase non linearity of the input stage, which means more distorsion.
Due to the Miller effect, the output impedance of the Vas main transistor is always much smaller than the CCS. Typically, it starts to decrease at 200 Hz. I've never seen a resistor in series with the CCS of the VAS, but, experimentally, it may not be a bad idea, at the expense of sacrifying the full voltage swing.
~~~~~~~~~Forr
§§§
pb said:
I suspect there is more to this than meets the eye....info on SPICE set-up...and models required....
In fact the output impedance of the ANF current source (my prefered name for this arrangement) is of the order of 500M-1G depending on the control elements current gain.....
Loop-transmission local to the current source ranges from 10dB to 40dB up to 200KHz.....
Stability margins are vast....typically >80 degrees....
The ANF current source's transient response is the equal of, and indeed superior to that of any other configuration...
Mikeks,
In ideal world ANF should be the best.
The SPICE is of course not a proof of anything. I actually was quite suprised to see any overshots in the very simple case:
see attached image comparing cascoded CCS and ANF.
I used LT SwitcherCAD and stock models.
The oscillation visible in ANF response decays within 10ns.
You might be right that there is more than meets they eye in that case (simulating engine artifact)
The pretty ugly stuff I have seen comes from, I admit, quite an extreme case: CCS for a class A single ended amp.
The power transistor was a MOSFET. When run with a more detailed SPICE model (I think it was one of IRFs with package inductances included) it would actually generate quite severe ringing.
This just proves that when parasitics are added in equation the behavior might be quite different.
BTW how do you know the phase margins? Care to share your results?
I was thinking of some more SPICE experiments but what really stops me is the knowledge that it may be a moot excercise since I have no access to a good quality models and simulator.
I guess we may need someone with network analyzer to build the real thing and measure it
Cheers,
Przemek
In ideal world ANF should be the best.
The SPICE is of course not a proof of anything. I actually was quite suprised to see any overshots in the very simple case:
see attached image comparing cascoded CCS and ANF.
I used LT SwitcherCAD and stock models.
The oscillation visible in ANF response decays within 10ns.
You might be right that there is more than meets they eye in that case (simulating engine artifact)
The pretty ugly stuff I have seen comes from, I admit, quite an extreme case: CCS for a class A single ended amp.
The power transistor was a MOSFET. When run with a more detailed SPICE model (I think it was one of IRFs with package inductances included) it would actually generate quite severe ringing.
This just proves that when parasitics are added in equation the behavior might be quite different.
BTW how do you know the phase margins? Care to share your results?
I was thinking of some more SPICE experiments but what really stops me is the knowledge that it may be a moot excercise since I have no access to a good quality models and simulator.
I guess we may need someone with network analyzer to build the real thing and measure it
Cheers,
Przemek
Attachments
Forr:
That's true unless VAS is a cascode. There is also a load of the output driver circuitry. Actually this is probably the lowest impedance of all (VAS out, CCS) and also prone to changes due to varied speaker impedance.
Say, three emiter follower stages with 100 hFE for each device each (quite optimistic) will give 1000,000 impedance multiplier. This will give 4-8Mohom load.
Might be interesting thing to check out, esp. where CCS operates with higher currents (larger transistor - higher capacitances). Also isn't the bootstrap another alternative?
That's true unless VAS is a cascode. There is also a load of the output driver circuitry. Actually this is probably the lowest impedance of all (VAS out, CCS) and also prone to changes due to varied speaker impedance.
Say, three emiter follower stages with 100 hFE for each device each (quite optimistic) will give 1000,000 impedance multiplier. This will give 4-8Mohom load.
Might be interesting thing to check out, esp. where CCS operates with higher currents (larger transistor - higher capacitances). Also isn't the bootstrap another alternative?
pb said:
As with most things, that depends on the application.....
Be that as it may, the ANF current source does appear to be the optimal arrangement for the second-stage (TIS) source in the generic topology....where its transient response, (given large, and very short inter-component PCB traces), is exemplary...
Its performance, of course, may be further enhanced by using fast-recovery Baker clamp diodes.....
PB
"That's true unless VAS is a cascode. There is also a load of the output driver circuitry. Actually this is probably the lowest impedance of all (VAS out, CCS) and also prone to changes due to varied speaker impedance."
Not sure exactly what happens because there is still the feedback effect by the Miller capacitor. The output impedance of the cascode is higher, so the -3dB frequency starts at a much lower frequency than in simple EC or CC + EC configurations for the VAS stage
"Say, three emiter follower stages with 100 hFE for each device each (quite optimistic) will give 1000,000 impedance multiplier. This will give 4-8Mohom load."
This is true at DC. Self recommands a CC buffer if a VAS cascode stage, otherwise the intended gain increase of the cascode is ruined. This effectively should give something like the impedance you mentionned but subject to the class B unlinearities...
"Might be interesting thing to check out, esp. where CCS operates with higher currents (larger transistor - higher capacitances). Also isn't the bootstrap another alternative?"
As far as I know the bootstrap (a very clever tehcnique) usually gives a bit more distorsion than a CCS but probably better symetry of slew-rate.
~~~~~~ Forr
§§§
"That's true unless VAS is a cascode. There is also a load of the output driver circuitry. Actually this is probably the lowest impedance of all (VAS out, CCS) and also prone to changes due to varied speaker impedance."
Not sure exactly what happens because there is still the feedback effect by the Miller capacitor. The output impedance of the cascode is higher, so the -3dB frequency starts at a much lower frequency than in simple EC or CC + EC configurations for the VAS stage
"Say, three emiter follower stages with 100 hFE for each device each (quite optimistic) will give 1000,000 impedance multiplier. This will give 4-8Mohom load."
This is true at DC. Self recommands a CC buffer if a VAS cascode stage, otherwise the intended gain increase of the cascode is ruined. This effectively should give something like the impedance you mentionned but subject to the class B unlinearities...
"Might be interesting thing to check out, esp. where CCS operates with higher currents (larger transistor - higher capacitances). Also isn't the bootstrap another alternative?"
As far as I know the bootstrap (a very clever tehcnique) usually gives a bit more distorsion than a CCS but probably better symetry of slew-rate.
~~~~~~ Forr
§§§
forr:
Oh, the intentionaly added Miller cap for frequency compensation.
I'm sorry I misunderstood you. I actually considered the circuit limits due to components, parasitics etc. before any compensation is added. Of course you are right - Miller cap as a neg feedback form will lower the VAS output impedance.
Yes, that's a trade off for a simpler circuit... and I was going to write "no feedback" but bootstrap is actually a positive feedback.
Well, now how about "sonics" of all those circuits? Is there a winner?
Or do we need 1MOhm resistor and 1kV power supply
Oh, the intentionaly added Miller cap for frequency compensation.
I'm sorry I misunderstood you. I actually considered the circuit limits due to components, parasitics etc. before any compensation is added. Of course you are right - Miller cap as a neg feedback form will lower the VAS output impedance.
Yes, that's a trade off for a simpler circuit... and I was going to write "no feedback" but bootstrap is actually a positive feedback.
Well, now how about "sonics" of all those circuits? Is there a winner?
Or do we need 1MOhm resistor and 1kV power supply
PB
"Well, now how about "sonics" of all those circuits? Is there a winner?"
There is too much inconsistency among subjective reports about the sonic virtues of such circuit or such other circuit that I prefer ot leave this kind of comments to other people.
I am very interested in the design of amplifiers and I try to detect the objective influence of every part of a given circuit. For me, the best stability and linearity are the main goals to reach.
~~~~~ Forr
§§§
"Well, now how about "sonics" of all those circuits? Is there a winner?"
There is too much inconsistency among subjective reports about the sonic virtues of such circuit or such other circuit that I prefer ot leave this kind of comments to other people.
I am very interested in the design of amplifiers and I try to detect the objective influence of every part of a given circuit. For me, the best stability and linearity are the main goals to reach.
~~~~~ Forr
§§§
Mikeks,
I wonder how important saturation protection circuitry.
This is one thing you notice comparing DIY projects to the commercial stuff where reliability matters a lot ( to bottom line )
Entering saturation will dramatically change the speed of the circuitry leading in a fast amplifier to oscillations, perhaps even destructive.
A good example is the Stochino amp found in this thread:
http://www.diyaudio.com/forums/showthread.php?postid=1000
It is not commercial but shows lots of attention devoted to clamping.
Note how VAS is clamped by reducing the voltage swing - Baker clamp.
Interesting also that ANF CCS for input dif pair use Miller cap.
Schematic.
http://www.soton.ac.uk/~apm3/diyaudio/Stochino_circuit.jpg
I wonder how important saturation protection circuitry.
This is one thing you notice comparing DIY projects to the commercial stuff where reliability matters a lot ( to bottom line
)Entering saturation will dramatically change the speed of the circuitry leading in a fast amplifier to oscillations, perhaps even destructive.
A good example is the Stochino amp found in this thread:
http://www.diyaudio.com/forums/showthread.php?postid=1000
It is not commercial but shows lots of attention devoted to clamping.
Note how VAS is clamped by reducing the voltage swing - Baker clamp.
Interesting also that ANF CCS for input dif pair use Miller cap.
Schematic.
http://www.soton.ac.uk/~apm3/diyaudio/Stochino_circuit.jpg
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.