Hi all,
Thanks for all the advice. No updates until now due to the usual - "Work is the curse of the drinking classes" - Oscar Wilde.
To answer several questions in no particular order - blmn, the oscillation is around 600 mV and is constant regardless of input. When I quickly measured it late last night I thought it was around 320 kHz - is actually 420 kHz.
I have found that by repeatedly connecting and disconnecting a 16ohm resistor test load that the oscillation will occasionally disappear when the load is disconnected. This is quite rare ( maybe 1 in 30/40 load removals ) and the amp will not oscillate until the load is reconnected.
Hugh, this amp design already contains 100R in the driver bases. The output stage is complementary feedback type (Sziklai-Pair) and I haven't yet tried the 10R resistors.
BAM - I will have to borrow a scanner so I might not get a schematic posted tonight.
Eric - I'll not try to summarise this article in case I miss some of the important detail, but for the above reason scans will take a couple of days. Output transistors are MJL3281A/MJL1302A BTW.
BAM - I have followed good psu and ground practices as you have described.
Thanks everyone,
James
Thanks for all the advice. No updates until now due to the usual - "Work is the curse of the drinking classes" - Oscar Wilde.
To answer several questions in no particular order - blmn, the oscillation is around 600 mV and is constant regardless of input. When I quickly measured it late last night I thought it was around 320 kHz - is actually 420 kHz.
I have found that by repeatedly connecting and disconnecting a 16ohm resistor test load that the oscillation will occasionally disappear when the load is disconnected. This is quite rare ( maybe 1 in 30/40 load removals ) and the amp will not oscillate until the load is reconnected.
Hugh, this amp design already contains 100R in the driver bases. The output stage is complementary feedback type (Sziklai-Pair) and I haven't yet tried the 10R resistors.
BAM - I will have to borrow a scanner so I might not get a schematic posted tonight.
Eric - I'll not try to summarise this article in case I miss some of the important detail, but for the above reason scans will take a couple of days. Output transistors are MJL3281A/MJL1302A BTW.
BAM - I have followed good psu and ground practices as you have described.
Thanks everyone,
James
--------------------
Hugh, this amp design already contains 100R in the driver bases. The output stage is complementary feedback type (Sziklai-Pair) and I haven't yet tried the 10R resistors.
--------------------
Please draw how you arrange the power supply rails, output devices and capacitors. Sounds like you have some parasitic inductance in supply rails. Unless you will stop the self generation, you will never get good sound, you will have a lot of in-band intermodulation products!
Hugh, this amp design already contains 100R in the driver bases. The output stage is complementary feedback type (Sziklai-Pair) and I haven't yet tried the 10R resistors.
--------------------
Please draw how you arrange the power supply rails, output devices and capacitors. Sounds like you have some parasitic inductance in supply rails. Unless you will stop the self generation, you will never get good sound, you will have a lot of in-band intermodulation products!
nemestra said:
Do you have and rail capacitance close to the amplifier itself? If so, do you have any capacitors directly across the rails (from + rail to - rail). If not try add several hundred uF from rail to rail right at the amplifier. (I had a similar problem once, and this seemed to cure it)
-Dan
Hi Nemestra,
The problem clearly is not in the output stage.
Suggest it might be the voltage amplifier, which leads us to compensation considerations, OR damping of the ubiquitous oscillations in the output stage.
The latter is scotch with 10R and 100nF in series from output to ground, and a 1.5uH choke in parallel with 10R at the output, OUTSIDE the feedback loop.
If the oscillation continues even after this treatment, next step would be around 15pF from output of VAS (collector is fine) to the feedback node. This pulls back OLG at very high frequencies and makes the output stage tolerant of capacitive loads. It's a favored technique of John Linsley Hood.
Cheers,
Hugh
The problem clearly is not in the output stage.
Suggest it might be the voltage amplifier, which leads us to compensation considerations, OR damping of the ubiquitous oscillations in the output stage.
The latter is scotch with 10R and 100nF in series from output to ground, and a 1.5uH choke in parallel with 10R at the output, OUTSIDE the feedback loop.
If the oscillation continues even after this treatment, next step would be around 15pF from output of VAS (collector is fine) to the feedback node. This pulls back OLG at very high frequencies and makes the output stage tolerant of capacitive loads. It's a favored technique of John Linsley Hood.
Cheers,
Hugh
dimitri said:
The first quote is not from me, but I can agree with the reasoning. Thanks for the Otala link!
I still think a) is correct. Consider an ihnerent Cbc that will vary with VCE from say 2-6 pF, a 300% variation. When you parallel a 22 pF, you only have 17% variation left, which is much better.
Eric,
On my commercial amp during development, I tried all sorts of compensation schemes. As a general rule, I found that with each successive reduction in lag comp from 100pF down to 36pF the sonics improved.
Below this, the sound worsened as the amp would lapse into short term oscillation on music.
I believe the theory relates to slew rate, which we could call the 'nimbleness' of the amplifier. As lag comp is reduced, so OLG increases at a given high frequency. As long as OLG gain is just below unity at the Bode frequency, all is well. This frequency is also determined by the load, the Zobel at the output, and the devices used throughout the amplifier, particularly the common emitter device.
On this basis, I searched through the design for ways of reducing OLG at very high frequencies without reverting to too much lag compensation of the VAS. I found one of the most useful was John Linsley Hood's technique of taking feedback directly from the collector of the VAS to the feedback node. This too has a bad effect on the sonics, but it makes the amp much more tolerant of capacitive loads - no bad thing as capacitive loads often precipitate amp instability anyway - since we invert the phase shift conferred by such loads with an interstage cap onto the feedback node. I also found that lag comp and interstage phase lead tend to act synergistically, making the amp easier to stabilize. The quality of these caps is also important.
Cheers,
Hugh
On my commercial amp during development, I tried all sorts of compensation schemes. As a general rule, I found that with each successive reduction in lag comp from 100pF down to 36pF the sonics improved.
Below this, the sound worsened as the amp would lapse into short term oscillation on music.
I believe the theory relates to slew rate, which we could call the 'nimbleness' of the amplifier. As lag comp is reduced, so OLG increases at a given high frequency. As long as OLG gain is just below unity at the Bode frequency, all is well. This frequency is also determined by the load, the Zobel at the output, and the devices used throughout the amplifier, particularly the common emitter device.
On this basis, I searched through the design for ways of reducing OLG at very high frequencies without reverting to too much lag compensation of the VAS. I found one of the most useful was John Linsley Hood's technique of taking feedback directly from the collector of the VAS to the feedback node. This too has a bad effect on the sonics, but it makes the amp much more tolerant of capacitive loads - no bad thing as capacitive loads often precipitate amp instability anyway - since we invert the phase shift conferred by such loads with an interstage cap onto the feedback node. I also found that lag comp and interstage phase lead tend to act synergistically, making the amp easier to stabilize. The quality of these caps is also important.
Cheers,
Hugh
capslock wrote--------
I still think a) is correct. Consider an ihnerent Cbc that will vary with VCE from say 2-6 pF, a 300% variation. When you parallel a 22 pF, you only have 17% variation left, which is much better.
---------------------------
Hi, capslock
Imagine an amp with inherent Cbc that will vary with VCE from say 2-6 pF. Assume it has closed loop gain 1000 at 20 kHz. The output variation will be 300%/1000=0.3%. Then you add 22 pF, the loop gain becomes 10 time lower. The output variation will be 17%/100=0.2%. The SAME value
You should read Cherry papers from JAES - very authoritative info
I still think a) is correct. Consider an ihnerent Cbc that will vary with VCE from say 2-6 pF, a 300% variation. When you parallel a 22 pF, you only have 17% variation left, which is much better.
---------------------------
Hi, capslock
Imagine an amp with inherent Cbc that will vary with VCE from say 2-6 pF. Assume it has closed loop gain 1000 at 20 kHz. The output variation will be 300%/1000=0.3%. Then you add 22 pF, the loop gain becomes 10 time lower. The output variation will be 17%/100=0.2%. The SAME value
You should read Cherry papers from JAES - very authoritative info
I was not talking about the variation in bandwidth, I was talking about nonlinear distortion due to change in capacitance with voltage swing. Reducing this variation by a factor of 15 will reduce the associated open loop distortion by a factor of 15. Bandwidth will also be reduced, causing lower feedback, but assuming you have to reduce it anyway, it may be better to choose the VAS rather than some other point to do the rolloff. This is all I was trying to say.
Latest
Hugh,
the Zobel networks you describe 10R+100n to gnd and a series 1.5uH || 10R are present. I have added a 22pF from the VAS collector to the feedback point ( base of long tailed input pair ) but no change.
I've borrowed a scanner but it's got a SCSI interface - it's going to be few more days before I can scan stuff - need to buy a SCSI card first.
James
Hugh,
the Zobel networks you describe 10R+100n to gnd and a series 1.5uH || 10R are present. I have added a 22pF from the VAS collector to the feedback point ( base of long tailed input pair ) but no change.
I've borrowed a scanner but it's got a SCSI interface - it's going to be few more days before I can scan stuff - need to buy a SCSI card first.
James
capslock said:
I've come to believe that miller compensation in large quantities is not a good thing. (I won't call it a bad thing... ...because sometimes it is necessary)
Listening tests have proved this... ...to myself and my wife that is
I'm not sure if it's the phase shift or the limiting of the slew rate of the amplifier that is the real cause. but, beyond a few tens of pF the sound becomes dulled and muddy. -Dan
capslock wrote------------
I was not talking about the variation in bandwidth, I was talking about nonlinear distortion due to change in capacitance with voltage swing. Reducing this variation by a factor of 15 will reduce the associated open loop distortion by a factor 15.
--------------------------------
I was also talking obout nonlinear disto. Your proposed method of reduction this variation by a factor of 15 will reduce the associated open loop distortion by a factor 15 and will also reduce loop gain by a factor 15, thus the distortion remain the same.
You start the thread, you ask questions and you don't like the answers that aren't coincide with your opinion - so what?
I was not talking about the variation in bandwidth, I was talking about nonlinear distortion due to change in capacitance with voltage swing. Reducing this variation by a factor of 15 will reduce the associated open loop distortion by a factor 15.
--------------------------------
I was also talking obout nonlinear disto. Your proposed method of reduction this variation by a factor of 15 will reduce the associated open loop distortion by a factor 15 and will also reduce loop gain by a factor 15, thus the distortion remain the same.
You start the thread, you ask questions and you don't like the answers that aren't coincide with your opinion - so what?
"I'm not sure if it's the phase shift or the limiting of the slew rate of the amplifier that is the real cause"
Can you think of any other causes?
dimitri:
You are assuming that the only impact is at low frequencies. The miller cap does reduce variation of amplitude and phase wrt output voltage. Negative feedback may mitigate this effect at low frequencies but will certainly not mitigate it at high frequencies where the gain margin is small. Is this important?
Can you think of any other causes?
dimitri:
You are assuming that the only impact is at low frequencies. The miller cap does reduce variation of amplitude and phase wrt output voltage. Negative feedback may mitigate this effect at low frequencies but will certainly not mitigate it at high frequencies where the gain margin is small. Is this important?
Hi, traderbam
The Miller cap does provide feedback around VAS, so VAS is in two feedback loops – in local Miller loop and in overall loop. Due to extra loop gain around VAS, the nonlinearity associated with VAS is effectively suppressed. The input stage and the output stage are only in the overall loop with low loop gain.
If the Miller cap is connected between the VAS output and the amp inverting input (Linsley Hood) the input stage will be in both loops the nonlinearity associated with input stage also will be effectively suppressed.
If the Miller cap is connected between the amp output and the VAS input (Edward Cherry) the output stage will be in both loops, and the nonlinearity associated with output stage also will be effectively suppressed.
--sorry for the hand drawing
The Miller cap does provide feedback around VAS, so VAS is in two feedback loops – in local Miller loop and in overall loop. Due to extra loop gain around VAS, the nonlinearity associated with VAS is effectively suppressed. The input stage and the output stage are only in the overall loop with low loop gain.
If the Miller cap is connected between the VAS output and the amp inverting input (Linsley Hood) the input stage will be in both loops the nonlinearity associated with input stage also will be effectively suppressed.
If the Miller cap is connected between the amp output and the VAS input (Edward Cherry) the output stage will be in both loops, and the nonlinearity associated with output stage also will be effectively suppressed.
--sorry for the hand drawing
Before you accuse me of not liking your answers, you should read carefully what I wrote. I was well aware of the reduction in open loop gain by a factor of 15. However, my point was that if you had the chance to roll off by the same amount either in the input stage or the VAS, maybe it would be advantageous to do so in the VAS. Doing it in the input stage does nothing about nonlinear distortion but reduces the open loop gain. Doing it at the VAS reduced loop gain by the same amount (or maybe even less, if advantage can be taken of pole splitting) while at the same time swamping out some of the C_bc nonlinearity.dimitri said:
capslock wrote ---------------
my point was that if you had the chance to roll off by the same amount either in the input stage or the VAS, maybe it would be advantageous to do so in the VAS
----------------------------------
You are right. How can we roll off in the input stage - only by the brutal force, to put a certain capacitance to the ground. We will get stability and nothing else. There are three ways to put Miller cap and to organize the extra inner feedback loop - see my previous post.
my point was that if you had the chance to roll off by the same amount either in the input stage or the VAS, maybe it would be advantageous to do so in the VAS
----------------------------------
You are right. How can we roll off in the input stage - only by the brutal force, to put a certain capacitance to the ground. We will get stability and nothing else. There are three ways to put Miller cap and to organize the extra inner feedback loop - see my previous post.
- Status
- Not open for further replies.