@RickRay
I’m not familiar with ARTA and find those measurements less straightforward to interpret than with REW.
It is now confirmed by proper measurements that denoiser in the VRDN regulator is not working to the max. potential in reducing the ripple. Noise is reduced by 30 dB but ripple is reduced by 18 dB. Interesting question, from the design perspective, is why is that?
I don’t think it is a question of Kelvin connection. Purpose of Kelvin connection is to avoid voltage drop along power carrying lanes introduced by their resistance and load current, and have exact output signal at the input of error amplifier. VRDN uses full available PCB width for power lanes and their resistance, at short distance between emitter of the denoiser transistor and output connector, is very small (70 uΩ), so voltage difference (error) is insignificant.
I’m leaning on strong capacitor charging pulses that induce voltage over the big area of ground plane, to be blamed. However, power planes instead narrow traces are necessary evil here as regulator is designed to provide 1.5 A output current.
I’m not familiar with ARTA and find those measurements less straightforward to interpret than with REW.
It is now confirmed by proper measurements that denoiser in the VRDN regulator is not working to the max. potential in reducing the ripple. Noise is reduced by 30 dB but ripple is reduced by 18 dB. Interesting question, from the design perspective, is why is that?
I don’t think it is a question of Kelvin connection. Purpose of Kelvin connection is to avoid voltage drop along power carrying lanes introduced by their resistance and load current, and have exact output signal at the input of error amplifier. VRDN uses full available PCB width for power lanes and their resistance, at short distance between emitter of the denoiser transistor and output connector, is very small (70 uΩ), so voltage difference (error) is insignificant.
I’m leaning on strong capacitor charging pulses that induce voltage over the big area of ground plane, to be blamed. However, power planes instead narrow traces are necessary evil here as regulator is designed to provide 1.5 A output current.
I did calculate my tht pcb trace resistance and came out at 6mOhm, but my actual board has solder all over that trace. I maybe should update the designs to keep the soldermask from the whole trace. Simple fix I think.
Anyway with a 6mOhm trace resistance at 1.5A that means that instead of 12Vout I'd have 11.991Vout.
But that would be without the cables + traces from the connectors of whatever you are powering to the actual load.
So if VRDN users want 1.5nV/sqrtHz level noisefloor and some extra PSRR, and some lower output impedance, they could use an add-on board with the VRDN. Applied correctly it would make for a really nice power supply.
Anyway with a 6mOhm trace resistance at 1.5A that means that instead of 12Vout I'd have 11.991Vout.
But that would be without the cables + traces from the connectors of whatever you are powering to the actual load.
So if VRDN users want 1.5nV/sqrtHz level noisefloor and some extra PSRR, and some lower output impedance, they could use an add-on board with the VRDN. Applied correctly it would make for a really nice power supply.
When you show these types of measurements you should change the vertical scale to focus on the meaningful information. In this case a range from 10pV to 1uV would be much better. It seems that your LNA (or environment) has lots of mains related noise.
The input ripple graphs of RickRay's VRDN (Measurement data and techniques for Elvee's De-Noizator: all implementations) and Hafler preamp (D-Noizator: a magic active noise canceller to retrofit & upgrade any 317-based V.Reg.) may indicate that there is some issue with VRDN.
VRDN input ripple
Hafler preamp input ripple
VRDN input ripple
Hafler preamp input ripple
If the LNA would have mains related noise, wouldn't it show up with its input shorted? That's from my desk, I mentioned the resistor wasn't shielded.
I also show mains related noise with a 1kohm resistor, but not with the leads shorted. It does not change the outcome. You can see from the picture how the add-on board was installed. The position of the sensing leads never changed, they were soldered to the leads of the 180 ohm load resistor.
The difference between the two readings isn't because of mains related noise.
The difference between the two readings isn't because of mains related noise.
No issue with VRDN measurement, I think there is an issue with the Hafler measurement. It is a simple LM3*7 with 1000uF filter caps before the regulators. No way it can be -60dB. I'll bet I forgot to change the multiplier for not using the LNA. So the real reading of input ripple for Hafler is more than likely 0dB, that would make way more sense in a lot of ways.
For one thing it would explain why I wasn't getting that much PSSR from the add-on board in that application.
I will re-measure the Hafler input ripple.
The 1k resistor noise I showed in VRDN thread was also made without any shielding (LNA's BNC input connector shorted with resistor). Maybe you should try to move the LNA to another location.
I replaced the two 0R39 resistors with 47R resistors. This moves the low-pass cut-off frequency from 185Hz to 1.5Hz. Of course there are trade-offs, the two new resistors now drop about 2 volts, so you would need a transformer with more voltage to reach +/-21 volts. It did decrease the input ripple from -40dB to -70dB which is more in line with the amount of caps on the front end of the VRDN.
Positive Rail input ripple with RC values changed
Positive Rail input ripple with RC values changed
I see what you are talking about. I went back and re-measured the Hafler and let it average 100 times. I get about -110dB@20Hz and the VRDN is at about -90dB@20Hz.
No idea what is causing that, like I said the Hafler has no extra filtering in the front end. Must be a mixture of components and implementation that created the nice low-pass response.
No idea what is causing that, like I said the Hafler has no extra filtering in the front end. Must be a mixture of components and implementation that created the nice low-pass response.
Haffler has smaller load current.
Ripple voltage at the capacitor can be calculated by using the following formula:
Vpp ripple = I / 2 * f * C
(use values as Volt Amp Hz Farad and f is mains frequency)
For instance, using that formula Haffler load current is 0.3 mA
Edit: corrected calculation results in too small current!
Are you sure it is not 10000 uF capacitor, or that ripple is really -60 dBV or just 1 mV?
Can you confirm that ripple by using oscilloscope?
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.