Ah for sure. I could only look in the audio spectrum off-course. I only have a DMM and the LNA that I built based on the previous posted schematic. There could be HF stuff that happens and I don't know about it.
This is just a hobby for me, and I don't think it's worth investing in better tools at this point.
I have been using Analog Device ADP and ADM series regulators for the last couple of years especially around digital i.c ..
To power the DAC I.C. since my first dac with TDA1541 I use only shunt regulators, they proved to be better in that position, now I put it for the PMD100 digital filter and so far it is the best reg I have used for it.
At 10MHz, a simple 1µF capacitor has a reactance of 16milliohm, which is probably lower than what a LTxxx or an ADxyz can achieve (and is vastly cheaper).
A less visible but more important attribute of a real passive is its ability to actually store and deliver energy. A regulator can only simulate it and can be limited by its internal slew-rate. It would be interesting to see the behavior of a high-perf reg when subjected to the max load chopped at 10MHz.
Even it it behaves gracefully itself, it will transfer the same demand to its input circuit which will have to be properly bypassed.
Naturally when discussing these regulators I'm not just referring to the regulator IC only but to the typical regulator application as specified by the datasheet. That includes an output capacitor which in these cases can be a low ESR type. You can check the performance of these regulator applications from the datasheet.
For power supply noise measurement you have to make sure that the amplifier's (LNA) input impedance doesn't unduly load the power source. For a sanity check, remember that once the noise of the amplifier is 3dB less than the anticipated noise of the source, the error will be less than 10%
You can parallel many ADA4898's as Gerhard has done, but Zin will be in the tens of Ohms.
Hewlett Packard got around this Zin issue by using positive feedback in the HP403 -- in the early years of transistors.
Folk take the FFT at face value -- but it really depends on the windowing. Measure the thermal noise of a 390 Ohm or 60 Ohm resistor with your LNA and the FFT -- it should be 2.5nV and 1nV respectively at room temperature (depending upon the RMS noise of the amplifier).
Another thing for the consideration is S/N ratio we can obtain in our amateurish conditions. This was clearly demonstrated on the VRDN measurement case.
Using adequate for the task LNA, RickRay was able to get clean fft graph with no 60 Hz or 120 Hz voltage, of resistor connected directly to the LNA input, at 4 nV/rtHz.
https://www.diyaudio.com/forums/att...el-ckts-11v-20v-1-5a-de-noiser-calibrated-jpg
So, we can reliably measure at least to that low level, right?
No, we can’t. Attaching measurement cables and connecting them to the DUT, changes situation. This is illustrated here:
https://www.diyaudio.com/forums/att...lvees-de-noizator-implementations-capture-png
For this, VRDN board was not connected to the transformer, otherwise 120 Hz ripple would be much higher. On that picture we see 60 Hz voltage to be at cca. -150 dBV (about 33 nV) as environment electromagnetic noise was inducing noise voltages in cables and DUT.
Convenient method for measuring PSRR, available to us average equipped diy’ers, is to use ratio between input and output ripple at the rectifier frequency (100 or 120 Hz).
However, if we have low ripple at the regulator input, let’s say 50 mV, at 100 dB PSRR, this ripple will be reduced to the 500 nV at the regulator output, and we are entering area of the EMI induced noise levels. Our measurement results would be compromised. That is why RickRay was able to measure only 95 dB PSRR at first, and around 105 dB when he increased input ripple voltage, thereby increasing S/N ratio. Real result is still above 110 dB.
Using adequate for the task LNA, RickRay was able to get clean fft graph with no 60 Hz or 120 Hz voltage, of resistor connected directly to the LNA input, at 4 nV/rtHz.
https://www.diyaudio.com/forums/att...el-ckts-11v-20v-1-5a-de-noiser-calibrated-jpg
So, we can reliably measure at least to that low level, right?
No, we can’t. Attaching measurement cables and connecting them to the DUT, changes situation. This is illustrated here:
https://www.diyaudio.com/forums/att...lvees-de-noizator-implementations-capture-png
For this, VRDN board was not connected to the transformer, otherwise 120 Hz ripple would be much higher. On that picture we see 60 Hz voltage to be at cca. -150 dBV (about 33 nV) as environment electromagnetic noise was inducing noise voltages in cables and DUT.
Convenient method for measuring PSRR, available to us average equipped diy’ers, is to use ratio between input and output ripple at the rectifier frequency (100 or 120 Hz).
However, if we have low ripple at the regulator input, let’s say 50 mV, at 100 dB PSRR, this ripple will be reduced to the 500 nV at the regulator output, and we are entering area of the EMI induced noise levels. Our measurement results would be compromised. That is why RickRay was able to measure only 95 dB PSRR at first, and around 105 dB when he increased input ripple voltage, thereby increasing S/N ratio. Real result is still above 110 dB.
Last edited:
But you confirmed his measurements on the stock VRDN board.
Also his second measurement with higher PSRR was not on a stock VRDN board, it was using an external bridge rectifier with a different grounding point on the PCB. That is not VRDN performance, but a VRDN with an external bridge connected in a different point, without CRC input filtering.
Also his second measurement with higher PSRR was not on a stock VRDN board, it was using an external bridge rectifier with a different grounding point on the PCB. That is not VRDN performance, but a VRDN with an external bridge connected in a different point, without CRC input filtering.
I think the input impedance of the LNA is around 8k. Which would be in parallel with the load resistor which was around 270R or something like that.
He also measured a 1k resistor for noise and is around 4.1nV/sqrtHz at 2-3kHz zone. And the noise of his LNA came out at around 0.46nV/sqrtHz at the same 2-3kHz.
No. The input impedance is the 10K of the bias resistor plus the input
impedance of the 20 op amps.
The input impedance of an ADA4898 is 30 Meg common mode
par 5K differential * loop gain, from the AD data sheet.
That is far away from tens of Ohms. There is no Zin issue other than the
input capacitance that starts to dominate above 10 KHz.
OK, maybe the need for a huge input coupling capacitor to short the noise
of the 10K through the low impedance DUT. If the DUT is not low impedance,
it will provide much more noise by itself than the amplifier.
The huge capacitor opens a completely new can of worms like
inrush current when connecting a DC voltage source.
There is an input impedance problem however with the usual FET amplifiers
that receive their feedback into the sources. The real part of the input imp.
turns negative above typically 70 KHz because with the feedback the FET
is no longer common source but more like a capacitively loaded follower.
Just plot re(generator_voltage/generator_current) for the input source
in LTspice. The negative input resistance turns the amplifier into an
oscillator depending on the signal source.
Just try your fave amp in LTspice.
cheers, Gerhard
Last edited:
My bad! Sorry Gerhard.
I re-read your 220pV/RtHz article and Zin is specified in a table.
Calibration in ARTA
Anyone here familiar with the calibration procedure in ARTA? I have a signal generator and am able to input 1vrms into the correct channel and perform the calibration process. When I do this, the graphs produced by ARTA directly display the correct values, for example a 1k resistor connected to the input of an LNA measure approx. -168dBV/sqrtHz and a 1 volt rms signal displays as 0dbV. What does the input sensitivity number that ARTA represent? ARTA calculated 4369.21mV when I input 1Vrms. I do not know if the number determined by ARTA is peak or p-p.
Anyone here familiar with the calibration procedure in ARTA? I have a signal generator and am able to input 1vrms into the correct channel and perform the calibration process. When I do this, the graphs produced by ARTA directly display the correct values, for example a 1k resistor connected to the input of an LNA measure approx. -168dBV/sqrtHz and a 1 volt rms signal displays as 0dbV. What does the input sensitivity number that ARTA represent? ARTA calculated 4369.21mV when I input 1Vrms. I do not know if the number determined by ARTA is peak or p-p.
Would like to understand this as well. So far I've used a value for input_sensitivity which when measuring the noise of a 1k resistor, the value in dBV/sqrtHz that it shows translates correctly applied to the input_sensitivity value (which I consider 0dB point) I have in ARTA.
ARTA manual shows that the dBV values are calculated based on the input_sensitivity value and preamp gain setting.
ARTA manual shows that the dBV values are calculated based on the input_sensitivity value and preamp gain setting.
Yes, RickRay is right, ARTA calibration value works as it does in REW, you calibrate and then read the dBV values relative to 0dB=1V, regardless of the calibration value in mV that it found.
I did manage to install REW in linux (forgot about java 8), and calibration spits out 1.47Vrms. But there's one issue in REW. If I add the LNA gain and I write 0.00147 and press enter, the 47 bit disappears and only 0.001 is left in the box. With a 1k resistor the noise is still around 4.02nV, just that if I alt-tab into another window, and come back, the noise drops to 2.8nV, so it does a kind of "refresh" and only reads the 0.001V value for calibration, and moves the graph down a bit.
So not really usable with adding the 60dB from the LNA, which is really a shame. Hope they fix this issue.
On another note, I used the value REW found, in ARTA, and seems to be working fine, as RickRay correctly was trying. The 1k resistor noise measures at -167.86dBV which with 0dB=1V is about right. With 1Vrms input signal ARTA shows it at around -1dB so that seems pretty good to me. I'll keep using ARTA, I don't like that gain thing in REW.
Because I used that different method of reading the measurements, the low level signals I measured lately should be in the same area.
My LNA noisefloor measured at -185.68dBV/sqrtHz at 2.5kHz which is around 0.52nV/sqrtHz.
I did manage to install REW in linux (forgot about java 8), and calibration spits out 1.47Vrms. But there's one issue in REW. If I add the LNA gain and I write 0.00147 and press enter, the 47 bit disappears and only 0.001 is left in the box. With a 1k resistor the noise is still around 4.02nV, just that if I alt-tab into another window, and come back, the noise drops to 2.8nV, so it does a kind of "refresh" and only reads the 0.001V value for calibration, and moves the graph down a bit.
So not really usable with adding the 60dB from the LNA, which is really a shame. Hope they fix this issue.
On another note, I used the value REW found, in ARTA, and seems to be working fine, as RickRay correctly was trying. The 1k resistor noise measures at -167.86dBV which with 0dB=1V is about right. With 1Vrms input signal ARTA shows it at around -1dB so that seems pretty good to me. I'll keep using ARTA, I don't like that gain thing in REW.
Because I used that different method of reading the measurements, the low level signals I measured lately should be in the same area.
My LNA noisefloor measured at -185.68dBV/sqrtHz at 2.5kHz which is around 0.52nV/sqrtHz.
Yes, it seems there is a bug in REW. I have noticed that REW rounds the entered "FS sine Vrms" number but had not noticed that REW actually uses the rounded number since in the soundcards I have used with REW the rounding error has been small (e.g. FS Vrms is 3,995V). I've sent feedback of this to REW developer.
I already received a fixed REW version from John Mulcahy (REW developer). I would expext a downloadable version to be available soon.
Just updated REW, had to recalibrate, a 1Vrms input signal was -20dBV down instead of 0dBV. After recalibrating, REW displays 34.52V as the "FS Sine Vrms" value. Correct value would be 3.452Vrms. Might need another update, decimal off one place to the right in some software line. Also explains why after updating it was -20dBV down.
VRDN Measurements
I have learned so much working with all these different boards, but have a lot more to learn I know.
I decided to implement an add-on board to the VRDN to see if I could increase the PSRR. The answer is yes. I believe the ground planes are picking up EMI that is out-of-phase with actual EMI and the out-of-phase EMI gets amplified and sent out.
By implementing an add-on board at the output of the VRDN I was able to gain 13dB of PSRR with a denoiser add-on board. The Dienoiser board added 17dB of PSRR and of course lowered the noise level.
I added the add-on board by connecting the V+/GND connections right at the output connectors. I removed the 220uF coupling capacitor and connected the adj. connection at the add-on board to the adj. pin of the LM317 where the capacitor was.
1kohm resistor noise measurement
VRDN stock Denoiser disengaged or Cadj. only measurements:
Noise@1kHz = -135.31dB or 171.5nV/sqrtHz
PSRR@120Hz = -83.6dB
Stock denoiser engaged measurements:
Noise@1kHz =-164.93dB or 5.6nV/sqrtHz
PSRR@120Hz = -100.99dB
Add-on denoiser implemented, stock denoiser not connected to coupling cap:
Noise@1kHz = -164.54dB or 5.9nV/sqrtHz
PSRR@120Hz = -114.08dB
Add-on Dienoiser implemented, stock denoiser not connected to coupling cap:
Noise@1kHz = -176.36 or 1.5nV/sqrtHz
PSRR@120Hz = -118.08dB
Measurement graphs in next posting.
I have learned so much working with all these different boards, but have a lot more to learn I know.
I decided to implement an add-on board to the VRDN to see if I could increase the PSRR. The answer is yes. I believe the ground planes are picking up EMI that is out-of-phase with actual EMI and the out-of-phase EMI gets amplified and sent out.
By implementing an add-on board at the output of the VRDN I was able to gain 13dB of PSRR with a denoiser add-on board. The Dienoiser board added 17dB of PSRR and of course lowered the noise level.
I added the add-on board by connecting the V+/GND connections right at the output connectors. I removed the 220uF coupling capacitor and connected the adj. connection at the add-on board to the adj. pin of the LM317 where the capacitor was.
1kohm resistor noise measurement
VRDN stock Denoiser disengaged or Cadj. only measurements:
Noise@1kHz = -135.31dB or 171.5nV/sqrtHz
PSRR@120Hz = -83.6dB
Stock denoiser engaged measurements:
Noise@1kHz =-164.93dB or 5.6nV/sqrtHz
PSRR@120Hz = -100.99dB
Add-on denoiser implemented, stock denoiser not connected to coupling cap:
Noise@1kHz = -164.54dB or 5.9nV/sqrtHz
PSRR@120Hz = -114.08dB
Add-on Dienoiser implemented, stock denoiser not connected to coupling cap:
Noise@1kHz = -176.36 or 1.5nV/sqrtHz
PSRR@120Hz = -118.08dB
Measurement graphs in next posting.
VRDN Measurement Graphs
All measurements taken on positive rail at 12volt output, 66mA load, boards upgraded to Rev. A2 except output capacitor changed to Panasonic FC 100uF/63V.
VRDN Positive Rail input Ripple
View attachment 929595
Positive Rail with stock denoiser disengage noise
View attachment 929596
Positive Rail with denoiser disengaged output ripple
View attachment 929597
Positive Rail with stock denoiser engaged noise
View attachment 929598
Positive Rail with stock denoiser engaged output ripple
View attachment 929599
Positive Rail Add-on denoiser implemented noise
View attachment 929600
Positive Rail Add-on denoiser implemented output ripple
View attachment 929601
Positive Rail Add-on Dienoiser implemented noise
View attachment 929602
Positive Rail Dienoiser implemented output ripple
View attachment 929603
All measurements taken on positive rail at 12volt output, 66mA load, boards upgraded to Rev. A2 except output capacitor changed to Panasonic FC 100uF/63V.
VRDN Positive Rail input Ripple
View attachment 929595
Positive Rail with stock denoiser disengage noise
View attachment 929596
Positive Rail with denoiser disengaged output ripple
View attachment 929597
Positive Rail with stock denoiser engaged noise
View attachment 929598
Positive Rail with stock denoiser engaged output ripple
View attachment 929599
Positive Rail Add-on denoiser implemented noise
View attachment 929600
Positive Rail Add-on denoiser implemented output ripple
View attachment 929601
Positive Rail Add-on Dienoiser implemented noise
View attachment 929602
Positive Rail Dienoiser implemented output ripple
View attachment 929603