Just LTSpiced a current mirrored LTP and I am seriously confused by the results.
Did a noise simulation with no emitter resistors for the current mirrors as a baseline.
Then added emitter resistors to the current mirrors because I wanted to see how much noise they would contribute - and the noise went DOWN!?
I expected that as I added resistence to counteract any small current mirror Vbe mismatch that the noise would increase, and that would be my trade-off.
Don't understand the actual result at all. After I read Self AND Cordell I expected I knew this stuff🙂. Anyone can help?
TIA David
Did a noise simulation with no emitter resistors for the current mirrors as a baseline.
Then added emitter resistors to the current mirrors because I wanted to see how much noise they would contribute - and the noise went DOWN!?
I expected that as I added resistence to counteract any small current mirror Vbe mismatch that the noise would increase, and that would be my trade-off.
Don't understand the actual result at all. After I read Self AND Cordell I expected I knew this stuff🙂. Anyone can help?
TIA David
the mirror degen R also serve as diff pair collector R giving Vgain R/gm, then mirror Q noise is added to the amplified V so it is a smaller contribution
alternatively the degen can be seen as reducing the mirror Q Vnoise conversion to current - both views should work out to the same net result
Bob's high diff pair degen values can be criticised on this issue, even larger mirror degen R is requrired to prevent 2nd stage noise contribuition
I haven't looked recently but Self likely treats this in more detail
but it is all trade-offs: input linearity vs noise, 2nd stage noise contribution vs Vheadroom, eventual excess forward gain pole with large mirror degen
alternatively the degen can be seen as reducing the mirror Q Vnoise conversion to current - both views should work out to the same net result
Bob's high diff pair degen values can be criticised on this issue, even larger mirror degen R is requrired to prevent 2nd stage noise contribuition
I haven't looked recently but Self likely treats this in more detail
but it is all trade-offs: input linearity vs noise, 2nd stage noise contribution vs Vheadroom, eventual excess forward gain pole with large mirror degen
Last edited:
Self only mentions the fact that it affects the slew rate, nothing on the noise issue. From simulation this can been seen but its a minor effect, I see using larger degen resistors on the current mirror to lower noise as a positive trade off where low noise is desired (pre amps) and in case of designs with very high LTP degenration (this offsets the slewrate problem) but in essence low noise in inputs of amps is a non issue, its already pretty low taking into account the line level voltages and input impedances. Samuel Groner wrote a paper about the noise issue and posted it on a thread here on diy. Noise can be substancially reduced through large degeneration of current mirrors.
Exactly my problem. His common emphasis on low impedance to minimize noise caused me to expect it applied here too, in the absence of any other comment.
This is to power some extremely efficient compression drivers, and in a bi-amp so they are not padded down to match the least efficient speaker unit. Hence about 25dBb more sensitive than typical. Thus my concern.
I need to think a bit more and then perhaps simulate before I know if I really understand but thank you both for the explanations and reference.
David
Please show the exact simulation setup. As jcx said, you need to look at the noise *referred to the input*, i.e. output noise divided by gain.
Samuel
Samuel
the HF channel will require 25dB less gain to match up the SPLs from the speakers.
> emphasis on low impedance to minimize noise
Low Z for desired sources.
High Z for un-desired sources.
Yes, the bias-sources probably should have large resistors to swamp voltage noise gain.
> to power some extremely efficient compression drivers, ... not padded down to match the least efficient speaker
Then reduce amplifier gain. S/N is established at the input. Say (random example) you have 1V of peak signal and 100uV of input hiss. 80db S/N. Say this should give 100db SPL on the room. Say the woofer needs 20V and the horn needs 2V. (Heavy woof for small size; hi-efficiency driver on hi-directivity horn.) The woof-amp could be set for gain of 20; horn-amp for gain of 2. The woof-amp's output hiss is 2mV and 20db SPL. The horn-amp's out hiss is 0.2mV and 20db SPL.
Problem: heavy-feedback amps get difficult at lower gains, when power device speed is not much-much-much greater than signal bandwidth. A low-tech amp (LM301) may give 10mHz GBW and slew 10V/us when compensated for gain-of-10, but must be slugged to 1MHz and 1V/uS when worked at gain-of-1.
OTOH, the few Watts needed for your horn allow "smaller" power devices which typically are available in higher speeds. Where a woofer may need 115-Watt blobs with 6MHz-30mHz fT, it may be practical to array 400MHz (TO-92) devices for a few Watts, close the loop for unity-gain at 10MHz (G=2 at 5MHz), therefore massive (>50db) NFB at the top of the audio band.
Low Z for desired sources.
High Z for un-desired sources.
Yes, the bias-sources probably should have large resistors to swamp voltage noise gain.
> to power some extremely efficient compression drivers, ... not padded down to match the least efficient speaker
Then reduce amplifier gain. S/N is established at the input. Say (random example) you have 1V of peak signal and 100uV of input hiss. 80db S/N. Say this should give 100db SPL on the room. Say the woofer needs 20V and the horn needs 2V. (Heavy woof for small size; hi-efficiency driver on hi-directivity horn.) The woof-amp could be set for gain of 20; horn-amp for gain of 2. The woof-amp's output hiss is 2mV and 20db SPL. The horn-amp's out hiss is 0.2mV and 20db SPL.
Problem: heavy-feedback amps get difficult at lower gains, when power device speed is not much-much-much greater than signal bandwidth. A low-tech amp (LM301) may give 10mHz GBW and slew 10V/us when compensated for gain-of-10, but must be slugged to 1MHz and 1V/uS when worked at gain-of-1.
OTOH, the few Watts needed for your horn allow "smaller" power devices which typically are available in higher speeds. Where a woofer may need 115-Watt blobs with 6MHz-30mHz fT, it may be practical to array 400MHz (TO-92) devices for a few Watts, close the loop for unity-gain at 10MHz (G=2 at 5MHz), therefore massive (>50db) NFB at the top of the audio band.
Just wanted some more information rather than an analysis of a particular circuit. And thanks to "homemodder" I found your excellent discussion of D. Self's book - that was exactly what I wanted so it's nice to be able say thank you directly. After I think about it a bit I may have questions about a particular circuit and I will definitely show the exact set-up if so.
Thanks
David
Last edited:
I expect I _will_ end up with a low gain amp but not 25dB less. As the next poster points out, it is not entirely simple to reduce the gain so much. And not only are there issues of stability but there are problems with the system gain structure. The amp won't match well with a normal active preamp and won't be usable for any other application.
So the lower the noise the more flexibility I have, and I wanted to learn more in any case.
Thanks
David
- Status
- Not open for further replies.