thank you for the test result, what does the impedance phase do? delay current?
Impedance curve and phase tell you under which conditions the circuit is stable.
With LTspice, think it is better done with a AC analysis with a AC 1 voltage source at the output.
In No 197 I show a design I might have made in 1973 if my tutor hade been a fan of Tobey Dinsdale or Mr HC Lin. One can add lets say a NE5534 to the output in a feedback loop. For many uses a resistor divider could be used. I did build that if I remember using a LED as bandgap. If memory is right it beat the TL431. Unlike the Jung circuit it was simple.
As far as I can see a LM317 is a LM741 ( 20 nV noise ) 1.25V bandgap, and simple output stage where we want a DC offset. A NE5532 in paralell should do 2nV. How quiet we can make a LED is something I forgot to measure( with capacitor ). Voltage stability wasn't bad. A Vbe multiplier could be tried. Often +/- 0.5V is OK.
LM723 is an interesting device. A bit slow which shouldn't matter if upstream decoupling is what it should be. Although it's output noise is higher than LM317 a simple additional capacitor makes it a winner. 10 uF already is doing a good job.
http://www.ti.com/lit/ds/symlink/lm723.pdf
As far as I can see a LM317 is a LM741 ( 20 nV noise ) 1.25V bandgap, and simple output stage where we want a DC offset. A NE5532 in paralell should do 2nV. How quiet we can make a LED is something I forgot to measure( with capacitor ). Voltage stability wasn't bad. A Vbe multiplier could be tried. Often +/- 0.5V is OK.
LM723 is an interesting device. A bit slow which shouldn't matter if upstream decoupling is what it should be. Although it's output noise is higher than LM317 a simple additional capacitor makes it a winner. 10 uF already is doing a good job.
http://www.ti.com/lit/ds/symlink/lm723.pdf
I didn't want to do a 10 K resistor. A Zinc Carbon 9V battery giving 9.9V was a surprise. No load so perhaps the best it could be. About the same as a LM7812 on a 1 amp load. Although the scope is near infinite input resistance the noise measurements improved with time( 5 mins )! Maybe even pico amps do something. This analyser hasn't been used in a long time. It was the one used in the past. I recalibrated it for the test.
This was the point all along. Test everything. This battery would still meet the needs of real world 24 bit. Although I can't say, some say Zinc batteries have the best sound. Perhaps the white noise is dither. Notice my LM317 was well off the graph.
I have given a pointer to this already above:
< LEDs | Voltage noise of some LEDs abused as references. 1K/1… | Flickr >
Avago HLMP6000 is below 1nV/rtHz above 100Hz. (1nV = 0 dB)
Someone on this BBS recommended that, but I don't remember who it was.
The circuit is just 14V from NiMH batteries, 2K resistor and LED.
HLMP6000 is cheap, available from Digi-Key IIRC.
Bandgaps are worse by design. After all, they do nothing but amplifying
the small voltage differences that result from different current densities
in pn junctions. They also amplify the noise.
Last edited:
The "non-invasive" method utilizes the relationship between the "Q" and phase margin:
Q=(√((Cos) φm))/sinφm
and Q = pi*f*T(g)
Comes from Erickson and Maksimovic Fundamentals of Power Electronics Springer 2004. Graphically explained in this application note around page 18:
https://www.omicron-lab.com/fileadm...ote_TraditionalNoninvasive_Stability_V1_3.pdf
My noise measurement amplifier is ~ 0.9 to 1.0 nV/RtHz. Gerhard's is 220pV/RtHz so he's your go-to guy!
Last edited:
But bandgaps are used because of the decent temp stability and easily implemented on many IC processes.
AFAIK LEDs V stability is all over the map basically , no better and most times worse than a pn junction. LEDs typical low Vf must be multiplied up too... Aren't you throwing out the baby to change the bath water?
OR how does stacked x3 Vf uncorrelated noise look against the time tested discrete temp compensated 6.2V zener? IMO unless you've got a Tcomp design in your back pocket, this is apples to oranges.
Last edited:
Bandgaps are easily 50 dB worse than a LED wrt noise.
The LT6655 is among the best, and with some filtering
it can be reduced to LED levels.
< lt6655_gimp | The LT6655 output noise gains a lot from an ou… | Flickr >
That may cost some DC precision due to op amp offset etc.
See the blue plot in the picture. The 50 Hz harmonics are there
because I did not have the time to bury everything in the multiple
aluminium shield boxes. You can ignore them.
Metrology grade precision is almost never needed in audio, 12
or 12.5 Volts usually does not matter if it stays where it is, but noise
can really hurt, especially with moving coil preamps that do not
have a PSSR worth speaking of.
Noise voltage of stacked devices adds geometrically, i.e. 4 LEDs
in series have only twice the noise voltage. Zeners get better
and better below 6V, when the avalanche component does no
longer play a significant role. Two BZX84-c3v3 will have _much_
less noise than one BZX84-C6V8, both for the stacking and for
the missing avalanche effect.
Multiplying the voltage of a low-voltage zener or LED is a bad
idea because the noise is also multiplied by N. When stacking,
it grows only with sqrt(N).
The bad thing is that the avalanche effect is needed for tempco,
so there are no compensated 3V3 diodes. For tempco, the sweet
spot is a little bit above 6V, depending on the process.
The LT6655 is among the best, and with some filtering
it can be reduced to LED levels.
< lt6655_gimp | The LT6655 output noise gains a lot from an ou… | Flickr >
That may cost some DC precision due to op amp offset etc.
See the blue plot in the picture. The 50 Hz harmonics are there
because I did not have the time to bury everything in the multiple
aluminium shield boxes. You can ignore them.
Metrology grade precision is almost never needed in audio, 12
or 12.5 Volts usually does not matter if it stays where it is, but noise
can really hurt, especially with moving coil preamps that do not
have a PSSR worth speaking of.
Noise voltage of stacked devices adds geometrically, i.e. 4 LEDs
in series have only twice the noise voltage. Zeners get better
and better below 6V, when the avalanche component does no
longer play a significant role. Two BZX84-c3v3 will have _much_
less noise than one BZX84-C6V8, both for the stacking and for
the missing avalanche effect.
Multiplying the voltage of a low-voltage zener or LED is a bad
idea because the noise is also multiplied by N. When stacking,
it grows only with sqrt(N).
The bad thing is that the avalanche effect is needed for tempco,
so there are no compensated 3V3 diodes. For tempco, the sweet
spot is a little bit above 6V, depending on the process.
ahh I see audio specialty designs, I reckon 90% of those would never pass a industry design review even with narrow environmentals. 12V? Doesn't 5-6V rails make sense for MC pre's*. If you are depending on noise from a particular LED I suggest you do a lifetime buy and do some incoming inspection., they change the process recipes quite often and that could bite you bad.
yes as a youngster I spotted the natural cancelling tempco effect of 6.x zeners.
FWIW There's some newer PLL / VCO Vregs that are worth a look. I forgot which ones. Hittite maybe. http://www.mouser.com/ds/2/609/hmc860-879491.pdf
Hard to imagine a MC pre would be worse, perhaps no neg feedback.
IDK perhaps use more moar active current sources..not fancy resistors
yes as a youngster I spotted the natural cancelling tempco effect of 6.x zeners.
FWIW There's some newer PLL / VCO Vregs that are worth a look. I forgot which ones. Hittite maybe. http://www.mouser.com/ds/2/609/hmc860-879491.pdf
Hard to imagine a MC pre would be worse, perhaps no neg feedback.
IDK perhaps use more moar active current sources..not fancy resistors
Last edited:
Oh, I use to survive design reviews, and I also do conduct them.
10 GBps Fiber optics, RF ultrasonics for reactor safety, and
next summer some of my stuff will fly to the space station.
Oscillators, power conditioners, timing circuits with ps resolution
and linearity.
My new low noise preamp _needs_ a 2nV/rtHz Vcc supply @ 10V,
despite having nice pssr. Below 100 pV/rtHz equivalent input noise
just everything hurts.
At what point does liquid nitrogen/helium and a vacuum break down in Boltzman? (I think that Scott mentioned something, but it completely escapes me...)
The LED graphs are impressive.
http://www.farnell.com/datasheets/1689832.pdf
http://cpc.farnell.com/stmicroelect...yIdBox=&selectedCategoryId=&iscrfnonsku=false
The LM723 is cheap and gives insights if wanting to clone it. Don't be put off by it's low gain after 1 kHz. That's just enough for the real world. The local decopling in the circuit can take over. No amount of PSU will fix bad decoupling.
http://www.farnell.com/datasheets/1689832.pdf
http://cpc.farnell.com/stmicroelect...yIdBox=&selectedCategoryId=&iscrfnonsku=false
The LM723 is cheap and gives insights if wanting to clone it. Don't be put off by it's low gain after 1 kHz. That's just enough for the real world. The local decopling in the circuit can take over. No amount of PSU will fix bad decoupling.
It's a big issue in RF, or for moving coil amplifiers.
Just some simulations -- will follow up on the bench --
pumped the LM317 10mA to 50mA, a 50uF 1uΩ ESR and 50uF 1Ω ESR output capacitor.
same with 10uF low ESR cap switched on and off the ADJ pin
Attachments
Last edited: