Hi Mike,
Referring to the TPC loop gain peak on page 179 of my book, and the use of a small bridging capacitor to mitigate it, I did another simulation of the issue with an amplifier having a single-transistor VAS and only an output stage double.
The TPC loop gain peak was not eliminated in the latter case, but it WAS reduced. Whereas the one on page 179 of my book had a TPC loop gain peak of about 20dB at about 1kHz, the second amplifier with a one-transistor VAS and output double exhibited a peak of about 10dB at 10kHz.
I do realize that this same peak can occur in the TMC minor loop if the same values of compensation network are used.
Cheers,
Bob
I always have this feeling.
"TMC is one pole compensation. Don't let bode plot fool you. That is 100% 1 pole compensation."
However, I do see 2 pole like frequency response. Even though, I don't believe it, either. There must be something wrong there. There must be some feedback CAP is feeding forward signal which is not supposed to do.(e.g. Miller Cap is very suspicious.). I plan to dive into this problem seriously.
"TMC is one pole compensation. Don't let bode plot fool you. That is 100% 1 pole compensation."
However, I do see 2 pole like frequency response. Even though, I don't believe it, either. There must be something wrong there. There must be some feedback CAP is feeding forward signal which is not supposed to do.(e.g. Miller Cap is very suspicious.). I plan to dive into this problem seriously.
Last edited:
This is quite logical but then it is right only as far as we look only
at this parameter , that is total loop gain devoted to VAS + OPS ,
for other parameters there will be a difference and that s why TMC
seems to me different from TPC.
It is true that TMC does not relieve the loading on the preceding stage like TPC does. As far as the input stage is concerned, TMC looks just like simple Miller compensation. So if your input stage is a significant source of distortion, TMC may not be the answer.
This is exactly right. TPC reduces the amplitude of the error at the input stage, TMC does not, when both are compared to straight Miller compensation. If the input stage is the dominant source of distortion in your amplifier, then TMC is not going to help. However, in well-designed amplifiers, the remaining dominant distortion that is most apt to hurt sound quality is in the output stage, for a variety of reasons, not the least of which is that it is often the only stage in the amplifier that is not operating in class A.
Cheers,
Bob
But if "TMC" does not improve on the loop gain in the audio band about the output stage compared with TPC, while TPC makes available the same extra loop gain to linearise both the input stage and the last two stages of the amplifier, of what utility is "TMC"?
Last edited:
Hi kgrlee,
I have run a sim below of the total loop gain enjoyed by the TIS and the output stage with "TMC" (green trace) by placing the probe in the innermost loop enclosing the output stage.
The blue trace is the major loop gain of TPC using the same compensation network values.
As you can see the loop gain in the audio band in both cases is exactly the same.
So, of what utility is "TMC" relative to TPC especially as the later provides the additional loop gain for the whole amplifier (including the input stage) while "TMC" does so for only the last two stages of the amplifier?
Attachments
Bob, I just thought, I addressed some misconceptions about current mirrors here:
http://www.diyaudio.com/forums/solid-state/192431-diyab-amp-honey-badger-19.html#post3312987
Maybe you could include similar information in your book.
A closely related issue is quasi-saturation in current mirrors. I think some attention to quasi-saturation is in order to dispel the stigma surrounding the use of BJTs in low-Vce situations. The right BJTs can be used to great effect at low Vce, such as in my K-multipliers. Using the wrong BJT in a Blameless amp for the LTP current mirror can sabotage TIS current gain, which is a crying shame when you've went through the effort to buffer it.
Some transistors can work usably well with Vce<600mV. One way to use this is to short the B-C junction of a transistor, then insert a schottkey in series with the collector. Then when current is passed through it, you have a 300mV reference voltage at the collector where the logarithmic transfers of the schottkey and transistor cancel out (depending on how well they match of course). If a resistor is used instead of the diode, and sized for 27mV voltage drop, the resistance will cancel the transistor's Rm (transresistance) and you will have a voltage source that will drop if Iq goes above or below the set point; this is useful for protection schemes.
http://www.diyaudio.com/forums/solid-state/192431-diyab-amp-honey-badger-19.html#post3312987
Maybe you could include similar information in your book.
A closely related issue is quasi-saturation in current mirrors. I think some attention to quasi-saturation is in order to dispel the stigma surrounding the use of BJTs in low-Vce situations. The right BJTs can be used to great effect at low Vce, such as in my K-multipliers. Using the wrong BJT in a Blameless amp for the LTP current mirror can sabotage TIS current gain, which is a crying shame when you've went through the effort to buffer it.
Some transistors can work usably well with Vce<600mV. One way to use this is to short the B-C junction of a transistor, then insert a schottkey in series with the collector. Then when current is passed through it, you have a 300mV reference voltage at the collector where the logarithmic transfers of the schottkey and transistor cancel out (depending on how well they match of course). If a resistor is used instead of the diode, and sized for 27mV voltage drop, the resistance will cancel the transistor's Rm (transresistance) and you will have a voltage source that will drop if Iq goes above or below the set point; this is useful for protection schemes.
Last edited:
TPC loads the VAS, which is a significant source of distortion in designs with heavily buffered output stages. If you have a heavily buffered output stage and a beefy input stage, TMC is superior as far as I recall.
I don't think TPC loads the TIS to any significant degree compared with "TMC", especially if the larger of the two capacitors is connected to the input of the TIS. Note that there is no such thing as "VAS" in the topology under discussion.
So, of what utility is "TMC" relative to TPC especially as the later provides the additional loop gain for the whole amplifier (including the input stage) while "TMC" does so for only the last two stages of the amplifier?
Another effect is that TPC will increase the input stage input Z ,
wich is useless , while TMC local bootstrapped feedback loop
will increase the VAS input Z hence reducing AC loading of the input stage
reducing its distorsion and increasing its gain if it is current mirror loaded.
When looking elsewhere than loop gain distribution there are other
differences that arise from the these two compensation schemes
even if final result is about the same in matter of perfs excepted
the higher slew rate of TPC compensated amps.
Actually, the input stage is a transadmittance amplifier, while the second stage is a transimpedance amplier. So the ideal impedance relationships between the two should be an infinite output impedance for the first stage and a zero input impedance for the second stage.
I fail to appreciate how "TMC" increases the TIS input impedance: both "TMC" and TPC are shunt applied negative feedback schemes, so they both reduce TIS input impedance.
Further, there is no way TPC can increase the input stage's input impedance because this is outside its local feedback loop.
Last edited:
It will increase it in the audio band , up to the frequency at wich
TPC and TMC start to have equal global feedback.
Below this frequency TPC has higher global feedback.
As you pointed it in the audio band TPC and TMC will give the same loop gain
about the VAS/TIS + OPS but not about the input stage , hence (input) impedance amplification at the said frequency is increased about the input stage.
With simulations you ll clearly see that depending on the compensation
the input stage AC input current is largely different , much lower with TPC.
Loop Gain & Tian probes
Bob, section 4.5 Evaluating Loop Gain could use an explanation & recommendation of Tian probes in LTspice
1st pic shows an amp with Inductor breaking the loop at -ve input of LTP as you suggest in the book. I sorta used this method for sims in Jurassic times. LGinductor.gif is the result.
As you say, the 'accuracy' of this depends on the difference in Z on either side of the inductor. If the Z at the base of Q3 is much greater than the Z at InN, we can expect the result to be OK.
baseQ3.gif shows the input Z at the base of Q3. 0dB is 1k and the input Z falls to that only above 27MHz.
We do the same for InN on the other side of the inductor in InN.gif
Ignoring the LF stuff due to the inductor, we see InN is sorta dominated by R9=600 as we expect and the Z there is always less than the input of Q3. It's also mostly resistive. We would expect little change to the Loop Gain at 10MHz and a bit more droop at 100MHz when we factor in the real Zs
As the Loop Gain goes to 0dB around 10MHz, we expect LGinductor.gif to be good enough to twiddle stability.
Lastly replacing the inductor with a Tian probe confirms this.
The Inductor is probably as good as the Tian probe in this example cos the 'large' difference between the two Zs.
However, if we probe other more sensitive parts of the circuit, the Zs might be less different or even crossover. Then the Tian probe calculates the effect of the different impedances for us. It's accurate & convenient.
Tian probes have other useful properties but these will do for us
Bob, section 4.5 Evaluating Loop Gain could use an explanation & recommendation of Tian probes in LTspice
1st pic shows an amp with Inductor breaking the loop at -ve input of LTP as you suggest in the book. I sorta used this method for sims in Jurassic times. LGinductor.gif is the result.
As you say, the 'accuracy' of this depends on the difference in Z on either side of the inductor. If the Z at the base of Q3 is much greater than the Z at InN, we can expect the result to be OK.
baseQ3.gif shows the input Z at the base of Q3. 0dB is 1k and the input Z falls to that only above 27MHz.
We do the same for InN on the other side of the inductor in InN.gif
Ignoring the LF stuff due to the inductor, we see InN is sorta dominated by R9=600 as we expect and the Z there is always less than the input of Q3. It's also mostly resistive. We would expect little change to the Loop Gain at 10MHz and a bit more droop at 100MHz when we factor in the real Zs
As the Loop Gain goes to 0dB around 10MHz, we expect LGinductor.gif to be good enough to twiddle stability.
Lastly replacing the inductor with a Tian probe confirms this.
The Inductor is probably as good as the Tian probe in this example cos the 'large' difference between the two Zs.
However, if we probe other more sensitive parts of the circuit, the Zs might be less different or even crossover. Then the Tian probe calculates the effect of the different impedances for us. It's accurate & convenient.
Tian probes have other useful properties but these will do for us
Attachments
Last edited:
The TMC phase at ULGF is better.
Plot is a bit small but it looks like an extra third of a radian or so.
And that is even with the TPC crossover frequency a little lower.
So there looks to be room to push up the feedback for the TMC and lower distortion a little.
Best wishes
David
Hi keantoken,
This is a good point, but I have not yet had time to fully absorb the other post. An ordinary 2-transistor current mirror nominally puts only one Vbe of Vce on the transistor since the base and collector of one transistor are connected together. I'm assuming that this is the low Vce condition you are referring to. If we get into quasi-saturation with this diode-connected transistor it can certainly cause degradation. This of course will depend on the current flowing and the type of transistor.
I usually use the "helpered" current mirror that is popular in integrated circuits, where an emitter follower is added to supply the base current of the two mirror transistors. This adds an extra Vbe of headroom to the transistor that is normally diode connected. I discuss this current mirror in Chapter 2 on page 35, Figure 2.9b. If this current mirror is used with a Darlington VAS, then both transistors of the current mirror see approximately the same 2Vbe Vce (not counting emitter degeneration drop). Does this address your point, or am I missing something?
Cheers,
Bob
TPC increases loop gain across the audio band. This means that the error voltage driving the forward path of the amplifier is significantly reduced compared with "TMC" or, for that matter, simple single-pole Miller compensation.
This in turn means that (with TPC) since the differential input voltage is reduced the input current will also be reduced for the same input voltage. This, I suspect, explains your observations, and does not in any way mean that the input impedance of the input stage has changed.
Last edited: