Emitter Current Noise Density

[Samuel Groner]

I think we (as often) actually agree but use different definitions. I still don't quite understand your use of SNR. If we have a unity gain amplifier with output/input noise floor of -100 dBu and use a 0 dBu signal, the SNR at the output (and input) is 100 dB. A 20 dB amplifier with output noise floor of -95 dBu has a SNR of 95 dB (5 dB less than before) for a 0 dBu output signal. If we refer the numbers to the input (-115 dBu noise and -20 dBu signal) SNR is still 95 dB, no?

Samuel

 

[JohnRoberts]

Our posts are crossing so the sequence may not make sense to a casual reader.

My mindset may be burned in from decades of low noise design where an implicit goal is to get circuitry noise floors down close to source noise floors. if you’re putting a gain stage in front of a distortion analyser presumably it’s to accomodate lower level sources. So the S/N realized by this source, or in distortion terms resolution floor is likewise important relative to the input.

I concede input S/N comparisons may not be fair since nominal input signal levels also routinely drop which neccesitated the increased gain in the first place. S/N relative to a fixed input level does indeed improve with increased gain but this only makes a difference when cascading gain sections and deciding how to parse gain between front end and later stage (like every console ever made). Input S/N without that input level output referenced will not be as meaningful.

I would be apprehensive about trying to design a SOTA distortion analyser these days in light of the current crop of new generation opamps. As it is these have to be run up to a few ten’s of dB noise gain to even see their distortion products with current bench equipment. Even with the increased gain the distortion products are lost below their own quite low noise floor up to a few volts output. We may need to look for some new plan of attack perhaps involving comb filters to look for harmonics preferentially over the noise. But this opens up another can of worms (A/D conversion, et al).

Yes, we basically agree and output S/N is the S/N that matters for the output.

JR

 

[bcarso]

Signal averaging can get you out of noise, although if the noise is high you may be averaging for a long time.

More insidious is distortion in the analyzer.

We also ought to start using some other signals than sinusoids, since they can conceal behavior that is potentially important, like dielectric absorption, self-heating effects, asymmetrical slewing second-order effects, etc.

 

[Samuel Groner]

:thumb:

I would be apprehensive about trying to design a SOTA distortion analyser these days in light of the current crop of new generation opamps.

SOTA = Society of Typographic Aficionados?

As it is these have to be run up to a few ten’s of dB noise gain to even see their distortion products with current bench equipment. Even with the increased gain the distortion products are lost below their own quite low noise floor up to a few volts output.

The (my) plan is to run the opamps at 40 dB or 60 dB noise gain and to use my fine Matlab skills for FFT analysis.

More insidious is distortion in the analyzer.

The notch filter per se seems not to be that difficult as you can implement it with inverting stages only and use precision passive components which allows to strongly decouple the required multipliers. However I'm really struggling to find a convincing way of implementing the input stage--100k ohm input impedance together with low noise and distortion seems like an unhappy combination. If we use noninverting gain stages we have high impedance and low noise but at least potentially higher distortion due to common mode effects. Inverting amplifiers are not low noise at 100k ohm impedance.

Samuel

 

[bcarso]

[quote author="Samuel Groner"] However I'm really struggling to find a convincing way of implementing the input stage--100k ohm input impedance together with low noise and distortion seems like an unhappy combination. If we use noninverting gain stages we have high impedance and low noise but at least potentially higher distortion due to common mode effects. Inverting amplifiers are not low noise at 100k ohm impedance.

Samuel[/quote]

Possibly look at the Funasaka-Kondou driven/bootstrapped supply approach for the unity-gain buffers.

 

[Samuel Groner]

Possibly look at the Funasaka-Kondou driven/bootstrapped supply approach for the unity-gain buffers.

Google refuses to give any further pointers on this--any reference?

[Edit: found it, AES preprint 1712.]

Samuel

 

[Samuel Groner]

US patent 4479093 seems to cover the basic idea.

Samuel

 

[bcarso]

Fascinating---three patents listed in this room in less than one day. Also of related note, Gus's thread linking to James Long's site.

Maybe Prodigy should have a special patent area. What they really say in the claims, what to worry about with regard to potential infringement, how to go about licensing, etc. How to search, whether it makes sense to file yourself, drawbacks of disclosure and maintenance fees...

 

[JohnRoberts]

[quote author="Samuel Groner"]:thumb:

I would be apprehensive about trying to design a SOTA distortion analyser these days in light of the current crop of new generation opamps.

SOTA = Society of Typographic Aficionados?[/quote]
I am sometimes thrown by the jargon around here and consider myself somewhat versed in the vernacular of engineering. I sprinkled in the SOTA trying desperately to fit in here. :wink:

The (my) plan is to run the opamps at 40 dB or 60 dB noise gain and to use my fine Matlab skills for FFT analysis.

This technique might downplay is input stage CM and dynamic range effects if tested inverting. At 100x or 1000x closed loop gain, even non-inverting the signal swing would be modest. Generating the high noise gain by shunting the feedback with a R from - input terminal to + input terminal will deliver high noise gain while properly exercising the input stage used non-inverted at modest overall closed loop gain. I suspect if that noise gain is set high enough it will interact with circuit's effective input impedance.

The notch filter per se seems not to be that difficult as you can implement it with inverting stages only and use precision passive components which allows to strongly decouple the required multipliers. However I'm really struggling to find a convincing way of implementing the input stage--100k ohm input impedance together with low noise and distortion seems like an unhappy combination. If we use noninverting gain stages we have high impedance and low noise but at least potentially higher distortion due to common mode effects. Inverting amplifiers are not low noise at 100k ohm impedance.

Samuel

A dedicated DOA can be made with decent input stage characteristics over a limited range. I still see some heavy lifting no matter what configuration to get this front end and full audio path better than the scary good potential candidates out there to test.

I'd be looking for some out of the box thinking, perhaps a more digital approach with unconventional (music sample?) test stimulus, and trick out a good codec to be great with some self calibration strategy. Perhaps learn the unique sample over N trials of looping the raw sample in. Of course evaluating the result would be a whole new exercise in trying to make meaning from chaos. Audiophools would eat that stuff up.

JR

 

[Samuel Groner]

This technique might downplay is input stage CM and dynamic range effects if tested inverting. At 100x or 1000x closed loop gain, even non-inverting the signal swing would be modest. Generating the high noise gain by shunting the feedback with a R from - input terminal to + input terminal will deliver high noise gain while properly exercising the input stage used non-inverted at modest overall closed loop gain.

Sure, W. Jung has documented this well. You might check my opamp test jig (OpAmp_Test_Jig_r1.pdf and opamp_test_jig.html).

A dedicated DOA can be made with decent input stage characteristics over a limited range.

CMRR is IMO/IME not an inherent forte of discrete design as it often relies on good transistor matching and excessive parts use (bootstrapped input cascoding and/or more complex topologies with CM feedback etc.). Of course one can use dual transistors to help things, but once in a while it becomes excessive at least with respect to cost. The bootstrapped supply looks like the more passable way right now.

I still see some heavy lifting no matter what configuration to get this front end and full audio path better than the scary good potential candidates out there to test.

It looks like distortion would residue at about -130 dB at 10 kHz for contemporary parts (AD797, LME49710) and unity gain, so at 40 dB noise gain this is -90 dB which is not too hard to get.

Samuel

 

[JohnRoberts]

I was under the impression some modern parts were better than -130dB but perhaps not at 10kHz. I don't follow the new stuff very closely. Indeed if all high linearity circuits can be tested at artificially high noise gains this will provide an adequate equipment margin.

Measuring full audio paths like consoles or power amps if very well executed might test the limits of a less intense design effort. To test our nails before we build our house perhaps, but not to test the finished house if engineered to an equally high standard.

JR