lumanauw said:Hi, Estuart,
Off course
Please submit them
I have 1 question. What happens if I don't use buffer before VAS (omitting Q2 in your graph, the input differential pair directly feds base of Q3)?
Hi lumanauw,
If Q2 is omitted, distortion of the VAS (2nd harmonic) itself raises to such a dominant level that improvements of the O/P stage are largely masked out.
Here some (simulated) figures:
THD at 10kHz and 30W
Traditional: 40ppm, w/o Q2: 120ppm
My trick: 6.2ppm. w/o Q2: 90ppm
Cheers
PS: I'll send you the schematic by e-mail, as it is too large to put it here.
Hi, Estuart,
OK, my email is lumanauw@bdg.centrin.net.id
Back to post#515
Why this method will not work with BJT output stage?
OK, my email is lumanauw@bdg.centrin.net.id
Back to post#515
Why this method will not work with BJT output stage?
Hi Andrew,
Low bias was used to study probs; years since I last did so.
Increase Iq to further reduce distortions (within reason).
The resistors linearise the phase response by loading a Miller connected C.dom and reducing output stage gain. Makes differential stage work harder and sounds better.
The split C.dom gives lower overall distortion and better load damping than using simple Miller C.dom which loads VAS more. Also this arrangement is not reactively tuned at a high audio frequency via a C-R connected to the output terminal and current phase shifting loudspeaker load.
However, there are so many ways for everyone to achieve desired results, with different choices relating to what individuals feel is most important.
Cheers ....... Graham.
Low bias was used to study probs; years since I last did so.
Increase Iq to further reduce distortions (within reason).
The resistors linearise the phase response by loading a Miller connected C.dom and reducing output stage gain. Makes differential stage work harder and sounds better.
The split C.dom gives lower overall distortion and better load damping than using simple Miller C.dom which loads VAS more. Also this arrangement is not reactively tuned at a high audio frequency via a C-R connected to the output terminal and current phase shifting loudspeaker load.
However, there are so many ways for everyone to achieve desired results, with different choices relating to what individuals feel is most important.
Cheers ....... Graham.
Bob Cordell said:
Hi Bob,
I fully agree with your comment. Connecting C2 to an emitter of a pre-driver does improve thd, but not much. However, as such arrangement might affect the stability of the inner loop, I didn’t dare to mention this possibility.
“Who originally came up with this form of compensation?” If you mean who first invented it, it seems to be Baxandall (according to Self). If you who mean who put it here, I must confess, I did.
Cheers,
mikeks said:If minor and major loop gains remain unchanged, then what's the point?
Putting the output stage inside the minor loop at audio frequencies, so its distortion is reduced by both the major and minor loops, instead of just the major loop. More constant (and less) distortion vs. frequency because the minor loop gain increases with increasing frequency in the audio range.
mikeks said:
I only looked at the major loop gain, and the network was the one I posted, rather than Edmond's. It was really just a quick check to see if the technique was worth pursuing. At any rate, the first sim I did showed big-time instability in transient, so I'm pretty sure it was a minor loop instability. The transient sim was also unstable when connecting a normal two-pole comp network to the amp output instead of the VAS output (as expected). I found that by making the resistor that connected to the output stage bigger and bigger, the system became stable, but I did not sim the loop gain of the inner loop.
mikeks said:Andy, it occurs to me that Edmond's arrangement here looks like a double-pole compensator with the resistor referred to the output instead of directly to ground.
It's hard to understand intuitively, because it's a three-port network. I don't pretend to have a good analytical grasp of it.
But I did think of this idea after a couple of beers last night
. In designing the two-pole comp, I use the simplified VAS transimpedance formula that has the two poles at the origin. It doesn't show the correct low freq behavior, but all I care about for design is where the zero is, and what the equivalent single-pole behavior looks like above the freq of the zero.One interesting way of calculating this formula is to treat the "tee" network as a general two-port. Suppose port 1 is the VAS input, and port 2 is the VAS output. Since the input voltage of the VAS is nearly zero, the transimpedance of the VAS can be estimated as the reciprocal of the y12 of the two-port. This assumes high minor loop gains. Computing the VAS transimpedance in this way results in the simplified formula with poles at the origin, and is easy to calculate. Now consider two two-port networks, one between VAS in and VAS out, and the second between VAS in and amplifier out. Over a pretty decent freq range, the VAS output and amp output voltages are almost equal. So for design purposes, it's as if these two two-ports are in parallel. For parallel two-ports, you just add the y12 values for each one to get the total y12. The reciprocal of this combined y12 should give a good approximation for the transimpedance of the VAS and output stage combo - as long as the frequency isn't too high.
Now if the y12 of the network from VAS in to amp out approaches a constant at infinite freq, and the y12 of the network from VAS in to VAS out approaches infinity at infinite freq, then the innermost network (VAS in to VAS out) will dominate the minor loop gain at tens of MHz where the unity loop gain frequency of the minor loop is.
This suggests that the following may be possible. Connect the two-pole comp network from VAS in to amplifier out. Except modify the network so there is a series resistor in one of the series arms of the tee. This will make its y12 approach a constant at infinite freq. The network from VAS in to VAS out could be just a single small-valued capacitor. The y12 of the cap will approach infinity as the freq does the same, so this innermost feedback path will dominate near the unity loop gain frequency of the inner loop, keeping the inner loop stable.
That's the idea anyway. I'm not positive it can be made to work, but it looks like a possibility.