Measuring low noise circuits

Hi,

What is the usual approach for measuring (the noise of) circuits which have a lower output noise than your analyser's residual? The easy route would be to put a very clean amp on the output of the circuit but this seems just too simple to me, what am I missing?

Thanks

with regards to noise only this is the common practice, yes. Why not? you must make sure that your preamp has no significant self noise / hum of course. excellent shielding, local battery powered operation is probably a must. balanced / single ended operation might be an issue. most measurement preamps I have seen are single ended.

some measurement systems can reduce noise through averaging, this might be a valid approach too.

- Michael

ej_whyte said:

Hi,

What is the usual approach for measuring (the noise of) circuits which have a lower output noise than your analyser's residual? The easy route would be to put a very clean amp on the output of the circuit but this seems just too simple to me, what am I missing?

Thanks

Click to expand...

Nothing...  While the added gain stage needs to have suitably low input noise.  I used to have a small battery powered +40dB fixed gain stage with noise weighting (IEC A) built into a small aluminum box. The "A" weighting roll-offs made the 40 dB gain a little easier for a simple transistor amp stage. The battery power made it easier to keep hum out of the measurement.

While I wouldn't trust the little bench box for published specifications it was great for measuring results while tweaking designs for noise. Even if not absolutely accurate, it will be useful for making relative comparison measurements. 

JR

+1 to everythibg that's been said so far.
Remember that the noise contribution of the additional preamp must be optimized not only by common techniques (operating point, shielding...) but also reducing its bandwidth to the domain of interest.
Most electronic noise is pink or white, so largely dependant on BW.
That was more or less implied in JR's use of an "A" weighting.

Another +1. An external amplifier is the standard way to do it. At Neve back in the 70s we had little die cast boxes containing a battery powered 40dB amp with weighting network. My present day Lindos audio test set has a built in 50dB amplifier, weighting network and quasi peak reading meter.

Just one word of warning. There are lots of different methods of converting the output of this amplifier into a reading representing noise in some way. There can be many dBs difference between the figures obtained from measuring the same piece of equipment using the different methods. I am sure you can guess which method the marketing guys prefer.

The attached file is the results of some tests done to reveal the differences between methods.

Cheers

Ian

ruffrecords said:

The attached file is the results of some tests done to reveal the differences between methods.

Click to expand...

Thanks for this Ian.  Do you have a date for it?

http://groupdiy.com/index.php?action=dlattach;topic=56276.0;attach=27753John Woodgate discounts the CCIR/ARM-2k measurement but in fact it made a lot of sense .. both for engineers as well as evil marketing.

Today however, the sensible way to do a meter is with some DSP in which case its easier to do RMS than it is to do Average Reading Meter.

ricardo said:

ruffrecords said:

The attached file is the results of some tests done to reveal the differences between methods.

Click to expand...

Thanks for this Ian.  Do you have a date for it?

http://groupdiy.com/index.php?action=dlattach;topic=56276.0;attach=27753John Woodgate discounts the CCIR/ARM-2k measurement but in fact it made a lot of sense .. both for engineers as well as evil marketing.

Today however, the sensible way to do a meter is with some DSP in which case its easier to do RMS than it is to do Average Reading Meter.

Click to expand...

I don't understand the final comment: "The fewer decimal places in the results, the more jittery the meter reading was."
Does it mean that the ballistics of the meter change with level? This is a known behaviour for moving-col meters with built-in rectifier because the motor's damping factor changes in accordance with the dynamic resistance of the rectifier, but I don't see how it applies to PPM's and RMS meters...?

abbey road d enfer said:

I don't understand the final comment: "The fewer decimal places in the results, the more jittery the meter reading was."

Click to expand...

I think he means that when the meter was jittery he found it difficult/impossible to determine a reading to as many decimal places as for cases where the reading was relatively steady.

Cheers

Ian

ruffrecords said:

abbey road d enfer said:

I don't understand the final comment: "The fewer decimal places in the results, the more jittery the meter reading was."

Click to expand...

I think he means that when the meter was jittery he found it difficult/impossible to determine a reading to as many decimal places as for cases where the reading was relatively steady.

Cheers

Ian

Click to expand...

Aah, OK, thanks.

ricardo said:

ruffrecords said:

The attached file is the results of some tests done to reveal the differences between methods.

Click to expand...

Thanks for this Ian.  Do you have a date for it?

http://groupdiy.com/index.php?action=dlattach;topic=56276.0;attach=27753John Woodgate discounts the CCIR/ARM-2k measurement but in fact it made a lot of sense .. both for engineers as well as evil marketing.

Today however, the sensible way to do a meter is with some DSP in which case its easier to do RMS than it is to do Average Reading Meter.

Click to expand...

I don't have a date for it. I have been having a long conversation on another group with Merlin from Neve about the measurement of noise in valves. It was he who found it and told me about it. It was in the context of trying to determine why his noise measurements were consistently better than mine. I was using my Lindos test set which measures to ITU-R 468 (originally CCIR- 1K). He was using an unweighted rms calibrated average reading meter IIRC. From the table, it looks like there could be as much as 8dB difference between the two.

It would be interesting to do a DSP noise meter which could be switched to make any of the various measurements.

Cheers

Ian

ruffrecords said:

It would be interesting to do a DSP noise meter which could be switched to make any of the various measurements.

Cheers

Ian

Click to expand...

I made myself a promise to not make hardware products that can be covered by a smart phone app. 

===

Back when I was messing with a microprocessor based console meter I experimented with computing RMS for the VU part of the Peak/VU display. On complex music with exact same release time constants I did not see a significant difference between RMS and simple average.

For noise measurements with higher crest factor perhaps.

Since square root is not a typical instruction in micros I had to roll my own software routine to extract the square root.

JR

OK good thank you all for the responses. Im using a dScope so very flexible options with regards to weighting, averaging, BW etc. and can already use the averaging to get a good look at very low THD.

I have a couple of AD797 lying around, does anyone have any quick links to a measurement amp circuit before I begin to roll my own?

Thanks

ej_whyte said:

OK good thank you all for the responses. Im using a dScope so very flexible options with regards to weighting, averaging, BW etc. and can already use the averaging to get a good look at very low THD.

I have a couple of AD797 lying around, does anyone have any quick links to a measurement amp circuit before I begin to roll my own?

Thanks

Click to expand...

If you want balanced, use the standard 3-opamp instrumentation amplifier. If unbalanced, just use a non-inverting stage. Just make sure the NFB path impedance is low enough.
For 40dB gain, I would use 10k and 100r (200r for balanced).

Anyone ever used or have thoughts on the Tangentsoft circuit?
http://tangentsoft.net/elec/lnmp/

I actually bought all the parts for that years ago and even did a board layout, but never got around to actually building it.

I wouldn't go for that, 100Ω input impedance, not good for low noise... Just a non inverting amp with 40dB should be good, if your tester has 0.1mV definition that would allow to measure with 1µV of definition. You could add another 20dB more gain in a switch if you want it. One pole is enough for decoupling DC. I would put a DC input too and a DC trim for low offset measurements and such, could be handy, but wouldn't trust much, just a little 8) could be much more trustable if bias current is used for the input transistors, as John Hardy mic pre, but too much tweaking and probably you would need something else to test it, don't know.

JS

mattamatta said:

Anyone ever used or have thoughts on the Tangentsoft circuit?
http://tangentsoft.net/elec/lnmp/

I actually bought all the parts for that years ago and even did a board layout, but never got around to actually building it.

Click to expand...

You need to be aware that this circuit has a very low input impedance (100 ohms), which makes it impractical for most measurements. As such, it makes the measurement sensitive to the impedance of the point that is actually measured. Even if you measure the output of a typical solid-state producr with an output Z of 50 ohms, the error will be more than 3 dB.
Due to the use of AD797, the VLN performance performance is guaranteed only for VLZ sources.
In particular, due to the inverting nature of the first stage, when nothing is connected at the input, the noise factor is worse than when connected to a 1 kohm source.

JohnRoberts said:

Since square root is not a typical instruction in micros I had to roll my own software routine to extract the square root.

JR

Click to expand...

Old thread, but this tidbit caught my attention, and may help someone working on a SW level meter, as I was a few months back:

If the end result is a dB calculation of an RMS signal, you can skip the square root of the RMS, and do it after the log() by dividing by 2 - the log of the square root of a number is one half of the log of the number. Yes, square roots are a PITA in a typical microprocessor or microcontroller. It's best to get rid them, if you can!

Regards,

Monte McGuire said:

JohnRoberts said:

Since square root is not a typical instruction in micros I had to roll my own software routine to extract the square root.

JR

Click to expand...

Old thread, but this tidbit caught my attention, and may help someone working on a SW level meter, as I was a few months back:

If the end result is a dB calculation of an RMS signal, you can skip the square root of the RMS, and do it after the log() by dividing by 2 - the log of the square root of a number is one half of the log of the number. Yes, square roots are a PITA in a typical microprocessor or microcontroller. It's best to get rid them, if you can!

Regards,

Click to expand...

If there was a machine instruction to perform a log conversion the square root would be a trivial one bit shift.. Working at machine level the math operations are more primitive, and even a simple divide takes multiple clock cycles.

I performed the square root computation using the fast (one clock tick) multiply to iteratively determine the square root one bit  per pass through. Since I did not need a full 16 bit result I could calculate the square root with several significant bits of resolution in a modest amount of clock cycles.

Indeed a look up table would have been more practical for a simple meter, but I amused myself by writing the algorithm (which is kind of what goes on inside a C library). 
=====
For another project I did a while back I needed to calculate actual dB for dynamic gain control and the log math characteristic where linear multiplies look like log addition can be used.  A dB ratio between two values can be determined again one bit at a time iteratively.  Oddly in that project I used a DPOT for the gain control element and they take linear control terms. But micros are pretty flexible if fast enough to perform conversions on the fly.


JR

Valley Audio used to make a noise level meter with a 100K input impedance, balanced input and selectable  filter settings. The front end was based on their Transamp circuit.
I think it was the model 310, a very handy tool for testing and evaluating noise.

BYacey said:

Valley Audio used to make a noise level meter with a 100K input impedance, balanced input and selectable  filter settings. The front end was based on their Transamp circuit.
I think it was the model 310, a very handy tool for testing and evaluating noise.

Click to expand...

Yup, Paul was knowledgeable and wrote articles about measuring input noise. Maybe 10-15 years earlier dbx made a cute decibel reading meter with a 2:1 companded front end... Odd looking VU meter with linear dB scale. Paul's later version had range switches and IIRC some extra features, The original dbx unit was very basic.

But these are obscure non-solutiuons for the OP's query...

Happy new year...

JR