I wonder how close yours is to my "Reasonable" effort I am working on?
I found TMC to greatly beat two-pole compensation. Darlington VAS the big key. No additional help from a cascode. Picking the exact correct parts and values is big. I think Bob and Douglas could have put a little more of why they picked some of their standing currents in their books. A lot came out as kind of SOP, but is that really right? I sure don't know enough tho challenge them as for every "fault" I find, there are ten more mitigating issues to be dealt with that make my assumptions to be wrong.
I know that someone needs to give this much more prominence.
For me CURRENT rules over almost all else.
Choosing the current, to me, seems to be critical to getting predicted performance.
Where did you get the 1.7nV/rt Hz from? there are many bipolar that can give lower noise than this, there is no theoretical limit except for the thermal noise from the source. For practical transistors there are limits depending on the rbb and hfe. I found this paper that gives some more information http://www.janascard.cz/PDF/Design of ultra low noise amplifiers.pdf
Learning some about that. I started with generic currents as commonly published, then decided they don't really smell right. Some calculation and some SPICE work, sure enough. I guess one needs to do a Gummel plot for every device to be sure you are in a reasonable region. Be sure you have "some order of magnitude" sufficient current to drive the next stage. That being dependent of the capacitance and peak current demands to find the BW. I have adjusted my design for a 5:1 to 10:1 ratio with the exception of the driver stage where that is just plain tough to do in a two-stage output. There are a lot of things either SPICE does not tell you, or I don't know how to ask it. It does not let you be totally lazy. So, balance between sufficient drive, sufficient speed and not too much as to increase noise. Again, I bet there are ten more things to consider.
It is the output. One has to enter the output node name before running the simulation. You can the plot that, or look at the contributions of individual components. But you can only view the noise voltage on the one node.
In LTspice, you specify a noise simulation. Run it, click on the output of your circuit and a graph of noise versus frequency will appear. It will most likely rise significantly at lower frequency, say below 100Hz. Now control click on the graph label. A little window will appear. Type in the stop and start frequencies. LTspice will then calculate the total RMS noise voltage for you.
As an example, I get 25.234uV for a circuit I am working on, for 20Hz to 20kHz. Peak output voltage is 34.6217V, or 24.28V RMS.
S/N ratio is 20 * log(24.28/25.23E-6) = 119dB. Not bad
CONTROL click. I would NEVER have found that! 15.57uV
So, filling in the blanks, -123 dB. Still, if it only much worse, it is great.
In fact, noise is going to be dominated by the preceding preamp in the vast majority of cases. A noise floor of 70 µV would be considered very good in an integrated amplifier. (Notoriously noisy ones may show 300 or even 500 µV.) If you do the math, at a total gain of 40-46 dB this corresponds to an input noise of .7 to .35 µV, or 5 to 2.5 nV/sqrt(Hz). So you're getting quite close to what's doable at reasonable effort.
The only ways of even getting close to power amp noise in practice would be either building an extremely low-noise preamp or a preamp with a two-stage volume control.
Of course you could also include an input gain control on the power amp, to be turned down when used with sensitive horn speakers and the like. Watch out for resulting source impedance though (power amps may exhibit considerable input impedance distortion when faced with high source impedance, due to input bias currents and Cob nonlinearity; this may be tackled by a bootstrapped cascode input). I would not go higher than 10k for the pot, which means 2.5 kOhms of source impedance max. If you say the latter still is too high and the former too low, I'd put low-noise opamp voltage followers on both sides (NJM2068 or LM4562 should be well-suited).
The only ways of even getting close to power amp noise in practice would be either building an extremely low-noise preamp or a preamp with a two-stage volume control.
Of course you could also include an input gain control on the power amp, to be turned down when used with sensitive horn speakers and the like. Watch out for resulting source impedance though (power amps may exhibit considerable input impedance distortion when faced with high source impedance, due to input bias currents and Cob nonlinearity; this may be tackled by a bootstrapped cascode input). I would not go higher than 10k for the pot, which means 2.5 kOhms of source impedance max. If you say the latter still is too high and the former too low, I'd put low-noise opamp voltage followers on both sides (NJM2068 or LM4562 should be well-suited).
Last edited:
Thank you so much for the kind help. Ok, I understand, however, one confusion is, for example. from here 160E-12 * SQRT(2.5E9), this 160 pv iis for a specific frequency of a frequency range. but we multiply with bandwidth not with that specific frequency, from where we get this 160 pV？
What about when this 160pV/rtH is at specific frequency 2.5Ghz. where the frequency range is from 600m to 2G. In that case what would be the noise amplitude? we got noise spectral density for a specific frequency, but we multiply with bandwidth? In that case, the noise amplitude is got for the whole bandwidth even noise spectral density was got from a specific frequency?
sorry for the confusion. noise density is defined or known for all frequencies. like 160pV for 2.5 Ghz, 290pV for 1.25 GHz, 480pV for 600 meg etc. Thst means I know all noise spectrum density for all frequnecies. Now, if I want to calculate noise amplitude, for example for 290pV, with which frequency I need to multiply? as specific one frequency and bandwidth are not same.
When you have different densities at different frequencies the calculation becomes an integration rather than a simple multiplication. I'm afraid doing integration is beyond my pay grade as math isn't a particular strong suit of mine. You could approximate the calculation into separate bands and then sum the noise powers as a way to avoid doing integration.
different densities for different frequencies. yeah, it is true. However, if I need to work with one density, means want to find noise amplitude for a specific frequency, then also simple multiplication wont work? need integration?
if I want to calculate noise amplitude just for one frequency, for example for 290pV density at 1.25G, with which frequency I need to multiply? with 1.25G as 290pv was got at this frequency? or with bandwidth? as specific one frequency and bandwidth are not same.
Thank you for introducing integration. however, to know noise amplitude for a specific frequency I am still confused.
sorry to bother....
if I want to calculate noise amplitude just for one frequency, for example for 290pV density at 1.25G, with which frequency I need to multiply? with 1.25G as 290pv was got at this frequency? or with bandwidth? as specific one frequency and bandwidth are not same.
Thank you for introducing integration. however, to know noise amplitude for a specific frequency I am still confused.
sorry to bother....
There cannot be 'a noise amplitude for a specific frequency' seeing as a specific frequency has a bandwidth of zero. There can only be noise in a particular finite bandwidth. So could you settle for noise amplitude per SQRT(unit bandwidth)? If so, then the density is multiplied by 1 to get that.
Oh I see. so even I get the noise spectrum density 160pV for 2.5G, multiply 160pV by sqrt(2.5G), hence 160pV * 50k = 8000000pV = 8uV is not working? as 2.5g is not bandwidth and also noise amplitude for a specific frequency not possible.
So, you suggest that "So could you settle for noise amplitude per SQRT(unit bandwidth)? If so, then the density is multiplied by 1 to get that. I am sorry i did not get fully, what would be unit bandwidth in that case and density is multiplied by 1?
sorry to bother you, but I need this help.
So, you suggest that "So could you settle for noise amplitude per SQRT(unit bandwidth)? If so, then the density is multiplied by 1 to get that. I am sorry i did not get fully, what would be unit bandwidth in that case and density is multiplied by 1?
sorry to bother you, but I need this help.
This calculation does not give the noise at 2.5GHz, rather it gives the noise in a 2.5GHz bandwidth i.e. 10kHz (say) - 2.5GHz.
It would be the noise in a unity bandwidth, i.e. per SQRT(1)Hz.