Johan,
You are correct to question simplistic SS current sinks in the tail of a differential phase splitter. A high AC impedance is needed. Cascoded FETs and BJTs work well. So does a pentode, but B- voltage has to go up.
Where loop NFB is applied to the non-inverting triode's grid, as is the case in 'El Cheapo", a good current sink is pretty much mandatory. The forced symmetry holds things together.
Hi Eli,
Watching the thread subject and your experience, this topic deviates somewhat yet doesn't necessitate a new thread. My further response:
.....and yet for some time it was almost standard practice in semiconductor topology to apply NFB via a differential pair without a CCS, rendering a few quite superb products. CCS only came later because it was simple to execute, and advantageous also because of the often notable spread in transistors. Point: how much bother is eliminated by the CCs compared to a resistor? (Darn, I don't have the spec analyser any more to work on this - retirement, cost and such
) On Spice this does not make a dramatic difference, other matters being well done. And twin tubes being well-matched .....
But again not criticising, just letting my thoughts go (politicians teach us not to do that). One of the merits of my own modest power amp design was to not use a diff. pair but mix NFB in passively, as per inverting op-amp style. (Cost me some effort to deal with resistor noise.) As per your closing remark; folks often seem to forget that any distortion introduced in the system of NFB mixing is not cancelled by NFB - downfall of more than one "good" product in the past.
You tempt me to go back and investigate further - unless I will be re-inventing the wheel. (Such a sobering experience: To have discovered/invented something, only to find that another already did that a decade ago.... )
Regards.
Watching the thread subject and your experience, this topic deviates somewhat yet doesn't necessitate a new thread. My further response:
.....and yet for some time it was almost standard practice in semiconductor topology to apply NFB via a differential pair without a CCS, rendering a few quite superb products. CCS only came later because it was simple to execute, and advantageous also because of the often notable spread in transistors. Point: how much bother is eliminated by the CCs compared to a resistor? (Darn, I don't have the spec analyser any more to work on this - retirement, cost and such
) On Spice this does not make a dramatic difference, other matters being well done. And twin tubes being well-matched .....But again not criticising, just letting my thoughts go (politicians teach us not to do that). One of the merits of my own modest power amp design was to not use a diff. pair but mix NFB in passively, as per inverting op-amp style. (Cost me some effort to deal with resistor noise.) As per your closing remark; folks often seem to forget that any distortion introduced in the system of NFB mixing is not cancelled by NFB - downfall of more than one "good" product in the past.
You tempt me to go back and investigate further - unless I will be re-inventing the wheel. (Such a sobering experience: To have discovered/invented something, only to find that another already did that a decade ago.... )
Regards.
You're right, I do mean something else. Most power amplifiers are configured to be non-inverting (just like op-amps). Alternatively, you could configure the entire amplifier to be inverting, allowing the feedback summing to be at a single point and avoiding the problem that Johan mentioned. The problem is that you are liable to end up with a 100k resistor in series with the input (sqrt{4kTBR}) and an even larger feedback resistor (capacitances become a problem).
Perhaps in some remote corner of this here website some folks will recognise daring when they see it....
To mix topologies for the moment: In my transistor power amplifier the input impedance is a nett 3,6K, and in the tube amp 7,5K. Noise as a result of that is only audible with the ear about 20 cm from the loudspeaker - a rather uncomfortable position in my case. Necessity then is an input buffer; for the transistor amp one with an output impedance of about 210 ohm and for the tube job a cathode follower drawing quite a few mA. Both buffers are included with the power amps, this also makes matters more independant of pre-amps and what.
Worthwhile?? A simulator uses an identical (in all ways) input pair; then not. But introduce differences such as may occur, and spectrum analysis does show lower high-order harmonics. I have not measured comparable results in the tube amp, as said before I do not find tube Spice models that useful (need perhaps a new programme, etc.)
There is also some slight advantage on overload, when things tend to go into cut-off somewhere. NFB lead/lag etc. - I am indebted to John Ellis for analysis and guidance in an article in EW on that.
To mix topologies for the moment: In my transistor power amplifier the input impedance is a nett 3,6K, and in the tube amp 7,5K. Noise as a result of that is only audible with the ear about 20 cm from the loudspeaker - a rather uncomfortable position in my case. Necessity then is an input buffer; for the transistor amp one with an output impedance of about 210 ohm and for the tube job a cathode follower drawing quite a few mA. Both buffers are included with the power amps, this also makes matters more independant of pre-amps and what.
Worthwhile?? A simulator uses an identical (in all ways) input pair; then not. But introduce differences such as may occur, and spectrum analysis does show lower high-order harmonics. I have not measured comparable results in the tube amp, as said before I do not find tube Spice models that useful (need perhaps a new programme, etc.)
There is also some slight advantage on overload, when things tend to go into cut-off somewhere. NFB lead/lag etc. - I am indebted to John Ellis for analysis and guidance in an article in EW on that.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.