I think he is just reading the cursor value of -171dBV at 20.5Hz
I'm wondering how to relate the numbers in red in the ARTA pane to the green spectrum profile. Looks to me like the green noise level is much lower than -143dBV (first pic in post 76)
Last edited:
But then noise/EMI on the balanced output cable shield pollutes the reference for the input, no? The noise profile I got when I had it setup this way was awful. Unfortunately I didn't save a pic of it and have subsequently replaced the standoffs with plastic ones.
Any view on post 78? I have a spare XLR connector here, but obviously there's no going back once I widen the hole on the front of the enclosure. Pin 2 XLR would be input signal (hot) and pin 3 input signal reference/GND. (Cable shield and pin 1 go to enclosure.) I'm using EMC series XLR connectors for the cable itself.
Any view on post 78? I have a spare XLR connector here, but obviously there's no going back once I widen the hole on the front of the enclosure. Pin 2 XLR would be input signal (hot) and pin 3 input signal reference/GND. (Cable shield and pin 1 go to enclosure.) I'm using EMC series XLR connectors for the cable itself.
-171dB is about 3nV/rtHz.
Was thinking that any reference I've seen so far about the BF862 mentions a noise corner frequency of hundreds of Hz. Example:
A low noise single-transistor transimpedance preamplifier for Fourier-transform mass spectrometry using a T feedback network
https://www.researchgate.net/public...ifier_for_Real-Time_Magnetic_Particle_Imaging
http://www.infn.it/thesis/PDF/getfile.php?filename=8817--specialistica.pdf
You got about one order of magnitude less. Perhaps the result of sorting the devices?
Why is your design not following at all the 1/4 power dependence of the noise density on the jfet drain current is yet another question. What SGK got (about -180dB at 1kHz) makes sense at low drain currents, you got much less around -190dB.
But never mind, I was just thinking out loud...
Perhaps a step closer. I switched out the BNC panel jack for another XLR. I then connected a balanced cable I had made up which uses the same cable (1m of Canare L-4E6S) I had used for the output connection to my balanced ADC and connector (Neutrik EMC series) and shorted pins 2 and 3 at the other end. At the moment the other end is still a female Neutrik connector and will need to be changed to crocodile clips or similar to be useful, but...
Attachments
Last edited:
The details are somewhat beyond what I can discuss here, but for best RF immunity you want to connect both input and output shield to chassis.
No! The scaling of this plot does not allow a direct reading of voltage noise density, see: http://www.diyaudio.com/forums/equipment-tools/287604-groners-low-noise-measurement-amp-linear-audio-vol-3-spare-boards-5.html#post4652886
No sorting (for 1/f noise) going on; just careful modeling of and attention to secondary noise sources which can become dominant at low frequencies if left unattended.
I'm not sure what you're referring to; I'm not aware that anything basic in my design is not explainable by standard theory.
Samuel
I'm definitely missing something here. If SGK measurement are not showing the noise density (which doesn't depend on the bin size) then how would you know his build reaches the target performance? ARTA or any other measurement software can easily plot the nose density vs. Frequency. What "scaling" do you mean? If not the input referred noise density, what represents -180dB on the graph?
You design runs the jfets at about 1mA. Though, the noise performance is very close to the jfets running at Idss (16mA). Theoretically yours should be two times worse, as 16^0.25. You mentioned something about in your paper, but still no clear explanation as of why this is not happening.
If not for 1/f noise, then why did you sort the jfets, as mentioned in the paper?
Well I now have that with the two XLR connectors and plastic standoffs. The shields of the input and output cables are connected to chassis (as is pin 1 of each of the XLR jacks). But the unbalanced input reference ("GND") is not and neither is GND on the board connected to the chassis. (If conductive standoffs are used then the board's GND reference is connected to chassis and the chassis, the board's GND reference, the single-ended input GND reference and the two cable shields with all their EMI are connected together.)
He has presented measurements in this thread where he compares the noise of the preamp with shorted input, and with a 1k resistor source. This allows a pretty good estimate of the voltage noise performance.
Both questions are clearly adressed in my article; I must presume you haven't read it?
So what is the unbalanced input reference connected to?
Samuel
Hi. Pin 3 of the XLR jack to J101 pin 2 (rightmost socket when looking from edge of board) i.e. circuit board GND. (Pin 2 of XLR jack goes to J101 pin 1. Pin 1 and shield to chassis.)
All I see here is that the preamp is some 20dB better in "something" compared to a 1k resistor. Doesn't mean anything to me regarding the noise performance, as specified in the design.
I wrongly assumed you are available for discussions, sorry for the trouble.
Last edited:
Seems to be ok. 
Attached is a measurement of my implementation off Douglas Self's low noise balanced to single-ended input board. The measurement was taken with the input board's inputs directly shorted rather than with a 50 Ohm resistor as Self used in his SSAD 2nd Ed. I will redo the measurement later with the resistor in place.
Are you really saying that little can be gleaned from the vertical scale readings?
If indeed the scale is true it would seem the board is measuring well in accordance with Self's discussion/measurements. (For now I only have the nominal 1000x amplification figure in the ARTA setup.)
I still don't understand ARTA's figures in red - a rough average across the frequency scale would suggest a performance much better than -88dBu
Attached is a measurement of my implementation off Douglas Self's low noise balanced to single-ended input board. The measurement was taken with the input board's inputs directly shorted rather than with a 50 Ohm resistor as Self used in his SSAD 2nd Ed. I will redo the measurement later with the resistor in place.
Are you really saying that little can be gleaned from the vertical scale readings?
If indeed the scale is true it would seem the board is measuring well in accordance with Self's discussion/measurements. (For now I only have the nominal 1000x amplification figure in the ARTA setup.)
I still don't understand ARTA's figures in red - a rough average across the frequency scale would suggest a performance much better than -88dBu
Attachments
You could use a 1000r as dummy input load as that has roughly the same noise as 4nV/rtHz. (I hope I have the 4nV right).
Then compare 1000r to 0r to see how far the plot moves on that vertical scale.
If I have this about right, then if the 0r goes down by 20dB relative to 1000r, the effective input noise is 20dB below that 4nV reference.
i.e. 0.4nV/rtHz
Then compare 1000r to 0r to see how far the plot moves on that vertical scale.
If I have this about right, then if the 0r goes down by 20dB relative to 1000r, the effective input noise is 20dB below that 4nV reference.
i.e. 0.4nV/rtHz
Last edited:
Yes, 1k is 4nV/rtHz. But the 10kHz noise is more like -17dB lower than the 1k value rendering .6nV/rtHz for the amp, which would make a lot of sense, also the 20Hz noise value would make (following the same logic) a lot of sense, it would be some 2nV/rtHz.
But Mr. Groner was reporting in Linear Audio completely different (much better) numbers for the performance (if memory serves, 0.35nV/rtHz @10kHz and 1nV/rtHz @ 10 Hz (that is about 0.7nV/rtHz @ 20Hz if the LF noise is flicker noise, therefore increasing by 10dB/decade) and I am not the one to prove him right or wrong.
It would so much easy to estimate the performance if SGK would plot his results straight in nV/rtHz on a log-log scale, each and every software can do this. And scale the graphs properly for reading the values, including the noise corner frequency.
I can plot whatever ARTA allows me to. In terms of scaling the options are dB FS, dBu, dBV and PSD.
The bit I am having trouble digesting is the following (I'm a relative newbie so bear with me)...
Apart from selecting the unit of measure as above, the figure placed in the ARTA setup menu (Setup -> Audio Devices) for pre-amp gain sets the scale. (If a power amp is involved this amp's gain needs to be added as well but for these purposes that field is left at 1). While I have for now entered the nominal value of 1000, we can test the pre-amp's gain for any frequency with a signal generator and scope (or true RMS MM). So if we know the amp gain, why is there any uncertainty over the scale?
Re the noise floor, there is of course both the noise of the pre-amp and the sound card.
The bit I am having trouble digesting is the following (I'm a relative newbie so bear with me)...
Apart from selecting the unit of measure as above, the figure placed in the ARTA setup menu (Setup -> Audio Devices) for pre-amp gain sets the scale. (If a power amp is involved this amp's gain needs to be added as well but for these purposes that field is left at 1). While I have for now entered the nominal value of 1000, we can test the pre-amp's gain for any frequency with a signal generator and scope (or true RMS MM). So if we know the amp gain, why is there any uncertainty over the scale?
Re the noise floor, there is of course both the noise of the pre-amp and the sound card.
PSD (Power Spectral Density), that's the right answer. You would also change the Y scale so that the graph fills the screen (like plot between -190dB and -160dB). Connect the 1k resistor to your amp input, not at the sound card input. Given the high gain of the amp (60dB) the sound card noise can easily be ignored.
I'm not sure why you keep mentioning the 1K resistor but of course it would be connected at the input of the measurement amp if I wanted to measure the noise of the resistor. I've not looked at measuring a 1K resistor since placing the preamp in its enclosure and finalising the cabling. If you would like me to measure a 1k resistor with this current test setup I can.
I attach the same measurement as before unchanged except to express the scale in dBV/sqrt(Hz) (by selecting PSD) and changing the range, although I'm not sure how it helps, to tighten it around the measurement of (a) my balanced to single-ended input board with its inputs shorted (green) and (b) the measurement I took of the noise floor of the test setup (yellow) with the input cable connectors shorted together. (b) is higher than recent posts because the alligator clips (and the micro clips I had to attach to them to measure the input board) seemed to pick up more EMI than when I had used a balanced cable and shorted the female end within the XLR connector. (For clarity, the measurement cable set up in use is XLR at the pre-amp end, with shield connected to the pre-amp enclosure etc as discussed above, and alligator clips at the other with an unterminated shield.) The peaks in the test setup noise floor to the far right of the displayed frequency spectrum seemed to be a moving feast - i.e. they came and went.
My ESI Juli@ XTe sound card is 192kHz capable. What FFT settings would you recommend for this sort of work? The variables are Hz (8k to 192k), FFT (4096 to 131,072), Window (Uniform, Hanning, Blackman3, Blackman4, Kaiser5, Kaiser7, FlatTop). I have been using 32 observation linear averaging.
I attach the same measurement as before unchanged except to express the scale in dBV/sqrt(Hz) (by selecting PSD) and changing the range, although I'm not sure how it helps, to tighten it around the measurement of (a) my balanced to single-ended input board with its inputs shorted (green) and (b) the measurement I took of the noise floor of the test setup (yellow) with the input cable connectors shorted together. (b) is higher than recent posts because the alligator clips (and the micro clips I had to attach to them to measure the input board) seemed to pick up more EMI than when I had used a balanced cable and shorted the female end within the XLR connector. (For clarity, the measurement cable set up in use is XLR at the pre-amp end, with shield connected to the pre-amp enclosure etc as discussed above, and alligator clips at the other with an unterminated shield.) The peaks in the test setup noise floor to the far right of the displayed frequency spectrum seemed to be a moving feast - i.e. they came and went.
My ESI Juli@ XTe sound card is 192kHz capable. What FFT settings would you recommend for this sort of work? The variables are Hz (8k to 192k), FFT (4096 to 131,072), Window (Uniform, Hanning, Blackman3, Blackman4, Kaiser5, Kaiser7, FlatTop). I have been using 32 observation linear averaging.
