Fair comment Waly.
The figures you mention seem to differ from Fig 14 in Samuel's paper. Am I missing something?
The figures you mention seem to differ from Fig 14 in Samuel's paper. Am I missing something?
The manual of my Agilent 89441A recommends a flat top window such as Hann
for accurate noise measurements.
There is a Hamming window, but no Hanning window. The guy was Justus von Hann,
from Austria.
Blackman windows have low sidelobes at -60 dB and down. That's nice if you
want to see things burried next to a clean carrier. Since you are measuring noise,
there are no empty spots @ -60 dB next to the loudest features where you could
find anything interesting.
With your soundcard, you could do a FFT from 0 to nearly 100 KHz. The 1/f region
would happen then in the first few bins. For the 89441A, I divided the range from
0.1 Hz to 1MHz into 7 decades, did one FFT per decade and combined the 7
traces into one with Gnuplot. That was a lot of software to write.
I have no idea if Arta can do something similar or if it only inflates the space for
the few bins on the low frequency side when drawing on a log. frequency scale.
Last edited:
Yes, sorry, you are right, I was quoting from memory. I checked the paper and Fig. 14 shows 0.45nV/rtHz at 20Hz (0.5nV/rtHz @ 10Hz) that's about half of what you got. About the same delta is in the noise corner frequency, where Mr. Groner reported less than 10Hz, which clashes with everything I have ever read or seen with jfets. Yours looks like it's in the low 100's of Hz, which is again what I would expect with jfets.
A few pages above I tried to ping Mr. Groner for a full explanation of his outstanding results, but got sorta cold shoulder. Based on what I quoted from the literature, I tend to believe your results are much more in line with what one should expect.
What I also find amazing in Fig. 14 is the lack of any mains frequency artifacts. Even air separated triple shielding of combined fero- and diamagnetic materials will still show some mains frequency peaking (and I'm telling from my own lab experience and other measurement results in the literature), unless you are doing the measurements with battery powered instruments in the middle of Mato Grosso jungle. In this respect, you got again what I would expect, which can be slightly further improved by better shielding.
The only result that aligns nicely is the 10KHz noise performance, you are very close to the reported result (0.45nV/rtHz vs. 0.38nV/rtHz reported).
Last edited:
Flat top windows are for maximum precision in the spectral components amplitude. If I'm not missing something, this also maps to precise values in the frequency bins, so it all makes sense.
P.S. With a sound card it's virtually impossible to go much under 20Hz. As a rule, the LF gain drops significantly under 20Hz, after all these are "audio" devices, not instrumentation.
Last edited:
Hi gerhard
Thanks for your input. The frequency response of the Juli@ sound card falls off a cliff above 22kHz. When I try to expand the frequency range of the FFT I get the same.
BTW according to Wiki, the Hann window is also known as Hanning window so perhaps we let ARTA off on that front.
Thanks for your input. The frequency response of the Juli@ sound card falls off a cliff above 22kHz. When I try to expand the frequency range of the FFT I get the same.
BTW according to Wiki, the Hann window is also known as Hanning window so perhaps we let ARTA off on that front.
Attachments
Last edited:
Tangled in GND think...
It is interesting to reflect on my "test lead" cable (the one that connects the pre-amp to the DUT) in light of section 4.5 of Bruno Putzey's Dealing with Legacy Pin 1 Problems article. The input to this pre-amp is single-ended and that section of the article deals with cabling single-ended sources to balanced inputs but bear with me for a bit.
For me the XLR jack on the pre-amp was one of convenience. With four conductor cable for my test lead I could connector one conductor pair to a red alligator clip and mark it "signal/hot". At the XLR end this pair connected to pin 2. The other conductor pair connect a black alligator clip marked "GND/cold" to pin 3. At the XLR/pre-amp end, the test lead shield is connected to pin 1 and pin 1 at the enclosure XLR jack is connected to the enclosure. At the alligator clip end the shield is unterminated. My idea here was that at least the shield formed some sort of Faraday cage over the conductors.
The XLR connector I am using is a Neutrik EMC series.
NC3MXX-EMC - Neutrik
It has a circular cap which connects the shield to the connector shell and a ferrite bead between the shield and pin 1.
I note that the first figure in sec 4.5 of the Putzey article has the shield connected to the GND reference conductor at the source end whereas mine is simply left unterminated. I wondered what the difference in noise measurement would be between using the test lead as I had set up versus connecting the shield at the DUT end to the clip probing the DUT's GND reference.
The measurement difference is very significant. See the attached pic which has two measurements of my balanced to single-ended input board. Yellow is my "DUT-end unterminated cable shield" version. Green is with the shield connected to the conductor pair probing the DUT ground reference - equivalent to the left side of Bruno's graphic. (The DUT ground plane is also connected to my bench power supply ground else the input board's ground is floating.)
What do you guys think is the more rigorous setup? Recall that my pre-amp board GND and battery supply have no direct connection to the pre-amp enclosure.
I guess I have effectively connected GND of the pre-amp to the pre-amp enclosure via the shield of the test lead (via a ferrite etc in the EMC XLR connector)...
It is interesting to reflect on my "test lead" cable (the one that connects the pre-amp to the DUT) in light of section 4.5 of Bruno Putzey's Dealing with Legacy Pin 1 Problems article. The input to this pre-amp is single-ended and that section of the article deals with cabling single-ended sources to balanced inputs but bear with me for a bit.
For me the XLR jack on the pre-amp was one of convenience. With four conductor cable for my test lead I could connector one conductor pair to a red alligator clip and mark it "signal/hot". At the XLR end this pair connected to pin 2. The other conductor pair connect a black alligator clip marked "GND/cold" to pin 3. At the XLR/pre-amp end, the test lead shield is connected to pin 1 and pin 1 at the enclosure XLR jack is connected to the enclosure. At the alligator clip end the shield is unterminated. My idea here was that at least the shield formed some sort of Faraday cage over the conductors.
The XLR connector I am using is a Neutrik EMC series.
NC3MXX-EMC - Neutrik
It has a circular cap which connects the shield to the connector shell and a ferrite bead between the shield and pin 1.
I note that the first figure in sec 4.5 of the Putzey article has the shield connected to the GND reference conductor at the source end whereas mine is simply left unterminated. I wondered what the difference in noise measurement would be between using the test lead as I had set up versus connecting the shield at the DUT end to the clip probing the DUT's GND reference.
The measurement difference is very significant. See the attached pic which has two measurements of my balanced to single-ended input board. Yellow is my "DUT-end unterminated cable shield" version. Green is with the shield connected to the conductor pair probing the DUT ground reference - equivalent to the left side of Bruno's graphic. (The DUT ground plane is also connected to my bench power supply ground else the input board's ground is floating.)
What do you guys think is the more rigorous setup? Recall that my pre-amp board GND and battery supply have no direct connection to the pre-amp enclosure.
I guess I have effectively connected GND of the pre-amp to the pre-amp enclosure via the shield of the test lead (via a ferrite etc in the EMC XLR connector)...
Attachments
Last edited:
Apologies for a long post. This is a stream of thought but bearing with me will hopefully help me untangle my 'ground think' problems.
Andrew I agree additional screening can only help as would a fully balanced eco-system, but my question is more about proper grounding techniques. (Bruno's G-Word article really resonates with respect to the absurdity of having a ground reference which is shared with all sorts of pollutants, but single-ended is a fact of life - unfortunately.)
Earlier in this thread, we discussed whether the measurement-amp board's GND should be connected to the pre-amp enclosure via its standoffs. I'm wondering if I have to revisit this. I had questioned this based on the noise I was seeing at the time and moved to plastic standoffs so that there was no direct connection. In contrast, Samuel had said "that's how it should be if you ask me" i.e. there ought to be a connection between signal ground and the enclosure of the pre-amp. I note also that Bruno's article shows such connection (I have circled it in pic 1) and in the text he says: "The audio circuit was also tied to the chassis, but crucially, elsewhere." Actually, this surprised me because I thought the whole point was not to commingle signal/audio ground and any currents circulating in the shields and enclosures. It sounds to me like I have missed a crucial point here and that I might have been wrong to move to plastic standoffs...
The measurement-amp is battery powered and I presume no distinction is made in the board between the battery pack ground reference and signal ground i.e. they both connect directly to the same plane. Does this change whether there ought to be a connection between measurement-amp signal ground and enclosure?
Balanced output connection from the measurement-amp is straightforward. Connect to XLR jack pin 2 and 3, ensure XLR pin 1 is connected to the enclosure and cable shield, transport signal and reference on separate conductors to a like setup at the sound card end (except in my case the balanced input to my ADC is via a TRS connector). Let's pinch one of the graphics in Bruno's article to depict this wiring. Pic 1. Easy peasy.
Now to the input side. The input to the measurement-amp is single-ended. The Source is the DUT and my DUT (my balanced to single-ended input board which also happens to have single-ended input on the same board) has single-ended output. But in many respects single-ended can be viewed as balanced where the reference signal just happens to be a constant. As Bruno's notes in his G-word paper, there's no need for the reference in a balanced connection to be symmetrically opposite to the signal - it can be anything, including some static value like "0V".
Think of a block diagram which simply shows an amplifier with its feedback resistors, call GND "reference" and show it wired directly from XLR pin 3 rather than as a bunch of GND symbol connections to a plane and there's little difference between single-ended input and balanced except that the reference leg will (be expected to) be constant 0v, rather than symmetrical with signal, and common mode rejection is destroyed. So how to connect a single-ended source to a "balanced" XLR connection? I looked at sec 4.5 of Dealing With Legacy Pin 1 Problems, reproduced the first graphic as pic 2 and tacked on the measurement amp block.
Cable and right side first. This is how I have it wired. What about the RC network? The Neutrik EMC connector has a circular capacitor from screen to connector shell and a ferrite bead between screen and pin 1. When the XLR connector meets the XLR panel jack and pin 1 of the panel jack is wired to the enclosure we effectively get an RC network. (Not the same values perhaps but the same purpose.)
Now the left side. I had left the shield unterminated at the DUT end. When I connected it to the single-ended DUT ground the noise dropped markedly. I was surprised by this. Perhaps I should not have been. If I were testing it en situe within the power amp enclosure there'd be a connection between this DUT signal ground to "star ground" and a connection from star ground to the enclosure. Had I placed an XLR "test lead jack" on the enclosure to make a convenient external connection to the measurement amp it would have connected the shield to the enclosure (which would in turn be connected to signal ground via star ground). Without it I could clip the test lead cable shield to the enclosure or simply connect it to the probe clipped to signal ground - as shown in the left side of Bruno's image. When testing the board without an enclosure I can only connect it directly to signal ground on the board (separately or via the ground probe) or to my power supply ground (they're both connected).
Were I to build a special purpose screening case as Andrew suggested then I'd be back thinking about whether the DUT signal ground should be connected to that case. If it were, Andrew's suggestion of clipping to the case amounts to much the same thing as clipping to signal ground. Of course, measuring the DUT while it is encased (either in its final enclosure or a screening case for test purposes) has the added benefit of less EMI ingress.
So should the measurement amp signal ground be connected to chassis? Should the 'test lead' shield connect to signal GND on the DUT? Seems the answers are likely "yes" to each but this ground think can get confusing as all hell. Any comments/thoughts would be appreciated.
Andrew I agree additional screening can only help as would a fully balanced eco-system, but my question is more about proper grounding techniques. (Bruno's G-Word article really resonates with respect to the absurdity of having a ground reference which is shared with all sorts of pollutants, but single-ended is a fact of life - unfortunately.)
Earlier in this thread, we discussed whether the measurement-amp board's GND should be connected to the pre-amp enclosure via its standoffs. I'm wondering if I have to revisit this. I had questioned this based on the noise I was seeing at the time and moved to plastic standoffs so that there was no direct connection. In contrast, Samuel had said "that's how it should be if you ask me" i.e. there ought to be a connection between signal ground and the enclosure of the pre-amp. I note also that Bruno's article shows such connection (I have circled it in pic 1) and in the text he says: "The audio circuit was also tied to the chassis, but crucially, elsewhere." Actually, this surprised me because I thought the whole point was not to commingle signal/audio ground and any currents circulating in the shields and enclosures. It sounds to me like I have missed a crucial point here and that I might have been wrong to move to plastic standoffs...
The measurement-amp is battery powered and I presume no distinction is made in the board between the battery pack ground reference and signal ground i.e. they both connect directly to the same plane. Does this change whether there ought to be a connection between measurement-amp signal ground and enclosure?
Balanced output connection from the measurement-amp is straightforward. Connect to XLR jack pin 2 and 3, ensure XLR pin 1 is connected to the enclosure and cable shield, transport signal and reference on separate conductors to a like setup at the sound card end (except in my case the balanced input to my ADC is via a TRS connector). Let's pinch one of the graphics in Bruno's article to depict this wiring. Pic 1. Easy peasy.
Now to the input side. The input to the measurement-amp is single-ended. The Source is the DUT and my DUT (my balanced to single-ended input board which also happens to have single-ended input on the same board) has single-ended output. But in many respects single-ended can be viewed as balanced where the reference signal just happens to be a constant. As Bruno's notes in his G-word paper, there's no need for the reference in a balanced connection to be symmetrically opposite to the signal - it can be anything, including some static value like "0V".
Think of a block diagram which simply shows an amplifier with its feedback resistors, call GND "reference" and show it wired directly from XLR pin 3 rather than as a bunch of GND symbol connections to a plane and there's little difference between single-ended input and balanced except that the reference leg will (be expected to) be constant 0v, rather than symmetrical with signal, and common mode rejection is destroyed. So how to connect a single-ended source to a "balanced" XLR connection? I looked at sec 4.5 of Dealing With Legacy Pin 1 Problems, reproduced the first graphic as pic 2 and tacked on the measurement amp block.
Cable and right side first. This is how I have it wired. What about the RC network? The Neutrik EMC connector has a circular capacitor from screen to connector shell and a ferrite bead between screen and pin 1. When the XLR connector meets the XLR panel jack and pin 1 of the panel jack is wired to the enclosure we effectively get an RC network. (Not the same values perhaps but the same purpose.)
Now the left side. I had left the shield unterminated at the DUT end. When I connected it to the single-ended DUT ground the noise dropped markedly. I was surprised by this. Perhaps I should not have been. If I were testing it en situe within the power amp enclosure there'd be a connection between this DUT signal ground to "star ground" and a connection from star ground to the enclosure. Had I placed an XLR "test lead jack" on the enclosure to make a convenient external connection to the measurement amp it would have connected the shield to the enclosure (which would in turn be connected to signal ground via star ground). Without it I could clip the test lead cable shield to the enclosure or simply connect it to the probe clipped to signal ground - as shown in the left side of Bruno's image. When testing the board without an enclosure I can only connect it directly to signal ground on the board (separately or via the ground probe) or to my power supply ground (they're both connected).
Were I to build a special purpose screening case as Andrew suggested then I'd be back thinking about whether the DUT signal ground should be connected to that case. If it were, Andrew's suggestion of clipping to the case amounts to much the same thing as clipping to signal ground. Of course, measuring the DUT while it is encased (either in its final enclosure or a screening case for test purposes) has the added benefit of less EMI ingress.
So should the measurement amp signal ground be connected to chassis? Should the 'test lead' shield connect to signal GND on the DUT? Seems the answers are likely "yes" to each but this ground think can get confusing as all hell. Any comments/thoughts would be appreciated.
Attachments
Anything connected to signal GND becomes part of the measuring circuit. Should I infer from your first paragraph that measurement-amplifier signal GND should not be connected to its enclosure (i.e. leave as it is currently)?
Only the output from the measurement amplifier is balanced. (Proper balanced connections require connection of enclosures.) The input is single-ended and the source/DUT is single-ended. The question is what's best/correct as a result.
Only the output from the measurement amplifier is balanced. (Proper balanced connections require connection of enclosures.) The input is single-ended and the source/DUT is single-ended. The question is what's best/correct as a result.
Last edited:
Proprer balanced impedance connections do not connect either of the signals to the enclosure.
The screen/shield is an extension of the enclosure. You end up with a dumb-bell enclosure where the two end bells are the cases and the middle hand grip is the cable screen.
Everything is inside the dumb-bell enclosure.
The smallest signal is at the input of the amplifier. It is that bit that should be balanced impedance for best intererence attenuation.
The output of the amplifier is 1000 times higher and benefits least from the balanced impedance connection.
The screen/shield is an extension of the enclosure. You end up with a dumb-bell enclosure where the two end bells are the cases and the middle hand grip is the cable screen.
Everything is inside the dumb-bell enclosure.
The smallest signal is at the input of the amplifier. It is that bit that should be balanced impedance for best intererence attenuation.
The output of the amplifier is 1000 times higher and benefits least from the balanced impedance connection.
That is what I thought and which drove the discussion some posts above (read from post 80) and which led to the switch from metal to plastic standoffs. And then I looked at the graphic in the Hypex paper which clearly shows - and talks about - an 'audio' signal ground connection to the chassis. Note also that moving to plastic standoffs effectively ground-lifted the measurement pre-amp (see section 4.2 of the Hypex paper).
Also, the input to the measurement amplifier isn't balanced, but rather it is single-ended. And unfortunately the power amplifier I am building - you are familiar with astx's SA2014 - is single-ended. Hence building Self's ultra low noise balanced to single-ended input circuit, and integrating single-ended into the single input board - to interface with it as the connections out of my DAC are only balanced. This board is the DUT of the moment.
So we have a mix of balanced and unbalanced and that's why I turned to the Hypex paper.
When I connect the test lead cable shield to signal ground of the DUT I get a noise measurement that is 10-15dB better than if I don't. (Pic in post 129.) Can this in any way be false positive?
Also, the input to the measurement amplifier isn't balanced, but rather it is single-ended. And unfortunately the power amplifier I am building - you are familiar with astx's SA2014 - is single-ended. Hence building Self's ultra low noise balanced to single-ended input circuit, and integrating single-ended into the single input board - to interface with it as the connections out of my DAC are only balanced. This board is the DUT of the moment.
So we have a mix of balanced and unbalanced and that's why I turned to the Hypex paper.
When I connect the test lead cable shield to signal ground of the DUT I get a noise measurement that is 10-15dB better than if I don't. (Pic in post 129.) Can this in any way be false positive?
Last edited:
Attached is how I would wire your setup using balanced XLR cables. Please note that I did not checked your PCB layout, so I'm only assuming it has a single common ground.
If I would have to lay out such a board, I would definitely separate the signal ground from the power supply ground, which would also change the wiring layout. But if your preamp is always battery powered, you should be able to get around with a single common ground as above, without a performance penalty.
Attachments
I'm not sure there is any compelling reason to ignore AES-48 with respect to the balanced connection from pre-amp output to sound card input. To my mind, the only real questions are (a) whether to connect the preamp board's GND to its enclosure or leave it floating (b) how to terminate the shield of DUT end of the input test lead when it has two conductors, one of which is used to probe DUT GND. Perhaps the two are not independent. Regarding (b) were the test lead a basic single-ended cable, i.e. single-conductor + shield, into a BNC connector it's simple. However, it would appear that better results can be obtained using two-conductor cable and another XLR jack rather than BNC.
If there is no chance of a "false positive", by which I mean a result that is better than reality because some real DUT noise is cancelled by the measurement setup, one can easily check the four possible combinations of (a) and (b) above...
If there is no chance of a "false positive", by which I mean a result that is better than reality because some real DUT noise is cancelled by the measurement setup, one can easily check the four possible combinations of (a) and (b) above...
Andrew suggested placing the DUT in a screening case and connecting the cable screen to the case.
I grabbed my metal Brabantia rubbish tin. Conveniently it has a lid that flips up and when the interior bucket is removed the base is open i.e. it forms a cylinder with a closable lid, and it has a circumference just big enough to fit over my balanced to single-ended input board I want to test. Simply, I slotted it over this single-ended DUT and measured again, firstly with the test lead screen still connected to DUT GND and secondly with the screen connected to the metal trash can. There was no connection from DUT GND to this makeshift screening case.
Unsurprisingly, having the shield helped a bit. The yellow trace in the pic attached is directly comparable to the green one in post 129 (note I extended the frequency range). But connecting the test lead screen to the screening case rather than DUT GND resulted in a much poorer noise measurement - perhaps because the single-ended DUT had no connection to its enclosure? This is why I worry about what I am calling "false positives".
I grabbed my metal Brabantia rubbish tin. Conveniently it has a lid that flips up and when the interior bucket is removed the base is open i.e. it forms a cylinder with a closable lid, and it has a circumference just big enough to fit over my balanced to single-ended input board I want to test. Simply, I slotted it over this single-ended DUT and measured again, firstly with the test lead screen still connected to DUT GND and secondly with the screen connected to the metal trash can. There was no connection from DUT GND to this makeshift screening case.
Unsurprisingly, having the shield helped a bit. The yellow trace in the pic attached is directly comparable to the green one in post 129 (note I extended the frequency range). But connecting the test lead screen to the screening case rather than DUT GND resulted in a much poorer noise measurement - perhaps because the single-ended DUT had no connection to its enclosure? This is why I worry about what I am calling "false positives".
Attachments
The penny finally dropped regarding this...
If the DUT has unity gain I presume no change is required as the pre-amp gain of 1000 is already factored into the results displayed.