nonsense

John Larkin's perceptions are predictably wrong. They let him see the world in way that makes him happy, but they don't do him any good.

Quite true, and if you are developing pro audio systems to get the best noise figure, a die cast box with internal partitioned batteries and screened output, is the only way to keep out all the rf and other noise found in a typical city environment.

Pro audio system developers are a mixed bunch. The difference between the objectivists and the subjectivists is considerable, and die cast boxes with internal partitions do look good. Electron beam microfabricators seemed to work perfectly well with less ostentatious screening.

They are, never a "golden ears" type here, but if you are trying to build the best possible noise figure input stages, the screened box approach is the only way to do it in a city environment. RF and power line noise gets in everywhere otherwise. Right at the ragged edge of available test gear performance, and they all ran on internal batteries.

We used to get transistors (pnp) already screened by TI, but then we would screen them again, using our own noise rig, to select the very best. There so much power gain in stage level pro audio systems, excessive hiss can destroy it for the front row and more.

Chris

A suitable FPGA would cost under $10. One might expect to spend a few dollars for a sub-PPM distortion variable-frequency sinewave generator.

I don't recall any realistic suggestions for measuring sub-PPM distortion.

"Use a spectrum analyzer" and "use a twin-tee to remove the fundamental" both ignore reality.

I've seen RF spectrum analyzers with -20 dB harmonic generation. Even a good audio range s.a. is likly to have ppms of internal distortion.

Maybe you can find an instrument that's better.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Might be easier to design and build a custom solution for frequency range of interest. So how would you approach that ?.

Bear in mind that at such levels, device and circuit noise might dominate the results, so perhaps there are practical measurement limits anyway.

Chris

I don't know. Even the best resistors have ppm/volt resistance change, and caps are much worse.

Yes, resistors make lots of noise.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

But still limited eventually, by the effective noise resistance of the input source.

Cambridge Instruments routinely put a grounded aluminium plate screen under every printed circuit board, until they moved on from 2-layer boards.

As soon as you can bury a more or less complete ground plane or power plane in a multi-layer board, an external screen isn't worth the effort.

The screened box had a long history, and kept on looking good long after it stopped serving any useful purpose,

Not if you know what you are doing.

And had gold-plated mains connectors.

And genuinely low noise op amps, like the LT1028 and the AD797 made discrete transistor front ends ostenatious confidence tricks, but they still worked as lures for gullible customers.

The NE5534 was cheaper, but good enough to wipe the floor with most discrete transistor designs, and in much less board space. 3.5nV/root Hz isn't as good as either the LT1028 (0.85nV/ root Hz) or the AD797.

That is what he just said.

Of course you don't. You didn't understand them, and you forget stuff that makes you feel stupid.

Your version of reality, in which you know what you are talking about.

The spectrum analyser's job is to reject signals that aren't at the frequency of interest. You end up looking at a very low amplitude signal in a very narrow frequency window. Ppms of internal distortion aren't going to create a problem there.

Why should I bother? You imagine some specious reason that would prove - to you - that it wasn't.

Actually they have missed what strikes me as an obvious solution. I may put in a provisional patent application to cover it. It adds less parallel capacitance to the inductor than any of the other solutions.

They are 10 times the price of an NE5534; you would only use them if the customer was prepared to pay a lot extra for a few nV improvement in the noise floor. Paralleling NE5534s might be more effective, depending on source resistance and how much less noise was actually required..

Paralleling amplifiers to reduce the noise voltage is useful only if the noise resistance (v_noise/i_noise) of the amplifiers is higher than the source resistance. In fact, you're trading voltage noise against current noise. The sweet spot is when both contribute equally: v_noise = i_noise * R_source. The match is not critical though.

It's still possible to do much better with discrete designs than with opamps. A noise voltage of 0.85nV/rtHz is not /really/ impressive, though for most audio sources it's plenty good.

In opamp circuits, the feedback network probably contributes more noise than the amplifier.

Jeroen Belleman

One of the biggest causes of a poor S/N ratio in badly-designed equipment is a terminating resistor across the input. If the correct terminating impedance can be arranged by a combination of series and parallel feedback, the S/N ratio will be a lot better.

(BBC AMC/5 was an example of a well-designed low-noise microphone amplifier.)

I've used a low-noise opamp with a unity-gain buffer inside the loop, so I could have a very low impedance feedback network and not overheat the good opamp.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

But how much noise did the unity-gain buffer contribute?

I've used transformer feedback instead of resistors. Of course, this only works when you don't need it to work down to DC. I used tiny amorphous metal cores with an amazing 13uH/turn^2, now unfortunately obsolete. Using those cores I could cover four decades of frequency.

Ferrites usually get you only three decades. That's still pretty good. When RF types say 'wideband', they mean anything over two octaves!

Jeroen Belleman

Very little, and its noise is essentially divided by the gain of the good amp.

This was amplifying a manganin current shunt for a 100-amp NMR pulsed gradient driver.

That's nice, for AC. Wrap a chopamp around that to go DC.

And to RF types, "DC" sometimes means a MHz.

Add "50 ohms" could mean 30. Close enough.

If you can set up a transmission line transformer you can get a few more octaves.