The simplistic Salas low voltage shunt regulator

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
diyAudio Chief Moderator
Joined 2002
Paid Member
As it is configured already. I.e. 1 positive and one negative per channel, same transformers. But its not best advised to ask specific modding steps here for a product we don't have experience with. The mods guy must figure it out. Maybe there are things or revisions there that we don't know about, and we can not take any responsibility by just looking at a web schematic. This is a diy thread specifically for the regulators, beyond that it carries a disclaimer.
 
R3?

Hi,

what is the purpose of R3. Earlier you suggested me, that I can short out this resistor. Without the resistor sound is mellower, warmer, but with resistor it is cleaner, colder but more agressive. So what is right?

Cheers!
 

Attachments

  • Salas_lv_shunt V1.jpg
    Salas_lv_shunt V1.jpg
    257.5 KB · Views: 556
quartz crystal based clock are an example, where you need good performance at 1 Hz,..10 Hz...100Hz, not good wideband performance.
On Walt Jung's website are great articles, "Regulators for High-Performance Audio", Part1, Part2 and Part3.
There are good examples, schematics and measurement of opamp based regulators and compared to discrete based regulators and popular 3-pin regulators like LM317, 7805,...

If someone would read those articles, he would see that discrete based regulator excels for wideband application, but opamp based regulators are better at low frequencies.

There are tons of noise studies for BJT, JFET and MOSFET available on internet. If someone would search & read those articles, he would see that BJT's are only choice for low 1/f (low freq) noise and MOSFETs are the worst choice.

And opamp differs too. Low voltage noise, low current noise is OK, how about low 1/f knee? And gain? If you crank up the gain, you pump-up the noise at output. Only thing that matters, is noise at output, therefore gain should remain low.

Example of good regulator for clock is Flea-power regulator

There are even better designs, but they were never published public and therefore they should remain intelectual property of their owners. Flea-power regulator is good starting point.

@Salas
I'm apologizing myself for OT and distraction.
 
Last edited:
diyAudio Chief Moderator
Joined 2002
Paid Member
Merlin uses MJE15031G output option and 220uF Vref filtering as I saw on his pic, recommended for clocks in the guide. Here is the output noise sim. Very low Rbb PNP follower, low noise BJT gain error amp 50dB OLG, heavily filtered passive resistive Vref, low noise Toshiba CCS JFETs. Check. :checked:
 

Attachments

  • BJTout+220uFatVref.gif
    BJTout+220uFatVref.gif
    7.2 KB · Views: 476
Merlin uses MJE15031G output option and 220uF Vref filtering as I saw on his pic, recommended for clocks in the guide.

Very good choice. But what is capacitive reactance of 220uF lytic at 1Hz or 10Hz? And CCS JFET's voltage noise?
Those simulated figures look very nice, but adleast few orders of magnitude too optimistic.
But only what matters is sound. If people are pleased with sound, this is only thing that matters.
 
diyAudio Chief Moderator
Joined 2002
Paid Member
Xc @ 10Hz = 72.343 Ohm. That is //~1kOhm in the 3.3V reg. The AD797 has 100 Ohm on its input pins, its bipolar so it has some Ib over them & over the gain ones, 2.58x gain via 1.3k/0.82k Rf in that link you provided. Those are altering the best the chip can do. There is no Vref gain in the BiB, only buffering. That shines more, the more Vout we need. I measured a Muse 220uF/25V for DF, has 0.0346 @ 100Hz, 0.0358 @ 120Hz, & 0.0925 @ 1kHz. So loss is normally dropping the lower we go. The 2SK117 Vref CCS is same league with AD797 (naked, best samples) for sub 100Hz near 1kOhm source and does much better near 10kOhm (Most people here will use ~5kOhm for 15Vout). There is no input bias current or other resistors on it. You don't get a 220uF Nichicon Muse in a SOIC package silicon die, nor 500mA idle current from its embedded output transistors. All in all not too bad for a discrete it looks like for elements and configurations, can have several nVrtHz 1/f in real life and not 1nVrtHz @ 20Hz, I don't buy that either, but its a good indication its not at least overlooked for noise as far as discrete go. And you can't ask say 45V @ 0.5A from an op amp, the BiB is general purpose. AD797 quotes 50nV p-p 10Hz BW self noise, forget the network around it. I give 30x simulator error because say the models are not good for complete noise calculations. If it gives 100nVp-p noise between 1Hz-10Hz in BJT output 3.3V guise, it would be nice. If someone with a BJT output BiB has a lab grade spectrum analyzer it will be nice to measure it for us. As for the subjective part, its up to each one's liking in the multitude of combinations it can serve. Some will hate it some will like it, I already provided the fairy tales part in the thread's title, so feel free shooting it or loving it and saying it here. Personally I will be glad if it can be adequate and useful in the hands of enough people in the long term if they will feel it was worth putting it together in the end instead of sticking an LM317.
 

Attachments

  • K117nf-f.gif
    K117nf-f.gif
    15.7 KB · Views: 543
  • AD797en-f.gif
    AD797en-f.gif
    6.8 KB · Views: 533
HERE is nice opamp noise calculator to make calculations easier and also some low noise calculations.

Xc@10Hz = 72.343 Ohm and Xc@1Hz = 723.43 Ohm for 220uF lytic cap. Therefore this cap won't do anything at low frequencies, since dynamic impedance of LED's is lower.

I'm great fan of discretes and prefer them over opamps. Yes, discretes give us more creativity, higher output voltages, more current,...etc, etc,...
Quartz based clocks are one funny exception, where we need low 1/f noise and discrete based regulator won't give us optimum performance. Especially JFET's and MOSFET's with higher 1/f corner. And clocks need only 3.3V or 5V.
 
diyAudio Chief Moderator
Joined 2002
Paid Member
No, there is an LED and a 1K trimmer in series. The cap will still do its stuff low enough since the trimmer will be at about 500R + LED impedance for 3mA running them. 1.9VLED at 3mA CCS what is it, 600 Ohm Z? Nice Calc BTW, tnx.

P.S. Isn't there any PSRR figure in the clocks? How sensitive they are?
 

Attachments

  • nvrtHzFlea.gif
    nvrtHzFlea.gif
    78.8 KB · Views: 530
Dynamic impedance of LED is not linear and Ohm's law doesn't work here.
Up to 1.7V, LED is usually OFF and resistance is very big.
From 1.7V to 2.0V LED is ON and ratio Voltage/Current is changing in non-linear way.
From 2V and up, LED is FULL ON and dynamic resistance remains constant.

Of course, all voltages are only approximation and can be different for different LED.

Dynamic impedance of single LED should be under 50R at 3mA. Correct answers can be found only in manufacturers datasheets.
 
diyAudio Chief Moderator
Joined 2002
Paid Member
Thanks for looking in to it. The LED's Z will come from DeltaV/DeltaI and charts I presume. OK, lets assume 50R. So we got a 550R Johnson noise source fed from a K117 CCS and shunted by a 220uF cap in our Merlin's example. So we must sum by square 3nVrtHz resistance noise and ~3nV rtHz @ 10Hz K117 noise. That's 4.24nVrtHz sans filtering. Though a 10Hz 72R Xc cap still affects a 550R Z even under 10Hz, lets say it has no effect. Still the fleapower AD797 arrangement gave 9nVrtHz assuming NO noise is coming in from its Vref which is an LED also with 10K and 3u3 filter, feeding 8.23dB gain. Can source 30mA. Not hands down easily better on second look.

P.S. 3.3uF=4.823k @ 10Hz & 48.3k @ 1Hz. I.e. much worse in proportion to its 10k than 72R to 550R and 720R to 550R.
 
Hmm, I'm sorry, but I don't quite understand your logic. Are we talking low wideband noise or low 1/f noise?
If we are talking about 1/f noise, we must compare apples to apples.

For AD797 noise from 0.1 - 10 Hz is specified as 50 nVp-p, that is 17.68 nVrms.

Where is specified 0.1 - 10 Hz noise for JFET?
If we assume there is 3nv@10 Hz, what is at 0.1 - 10 Hz?
30 nV rms....50 nVrms...70 nVrms.....?

In BiB regulator noise from voltage reference (LED + 500R trimmer) is shunted by 220uF capacitor (bigger would be even better) + noise contribution from CCS. This is feed through R104 to base of Q104. Q104 is BC550 with base spreading resistance of 150R@3mA and 145R@5mA. And there is another CCS, Q105.

In Fleapower regulator situation is not that bad. 10K and 3.3uF are forming LPF with Fc(-3dB) @ 4.82 Hz thus filtering voltage reference noise and shunting noise to GND (I would filter even lower, bellow 1 Hz). Shunting capacitor across 1K3 feedback resistor is decreasing AC gain (noise gain).

Anyway, I never stated that BiB is bad and Fleapower is SOTA regulator. I only gave Fleapower as start point for low 1/f noise regulator suitable for clocks.
 
diyAudio Chief Moderator
Joined 2002
Paid Member
550R and 220uF is 1.32Hz, the cap is supply rail referenced. More than 220uF does not do much for total noise by running iterations up to 1000uF in noise sims, just slows the CCS regulation on the resistor. Maybe due to other noise sources prevailing. But for clocks 10Hz under noise, a 1000uF can be fitted no problem, there is space. They don't give noise density unfortunately for the Toshiba, just a 10Hz end dB value for 500R, 1K, 10k in the diagram I posted. Its almost equivalent to a K170 noise wise, which is very popular and maybe there is more data if we look around. I was just looking at components in each one and their possible noise generation for studying them. There is a 7812 CCS leg and output rail in the Flea to consider, BC547 BJT pass element final version in page two flea, same Rbb as BC550, as there is also a main CCS in the 1.1 etc. I don't have clocks and spectrum analyzers to fit both and see how the clocks pick up, if there is PSRR in clocks resisting some noise etc. I would make it out of curiosity since I have the parts and LT1028 even. Its just good learning to think by looking in specialized circuits, I never designed for clocks, or I especially recommended anything I made for digital. But its interesting discussion. I still give a fat order noise error and I estimate the BJT 1.1 3.3V at 100nV in the last 10Hz octave. And I am sure that a flea will multiply the basic 797 noise to a serious extent too.
 
diyAudio Chief Moderator
Joined 2002
Paid Member
*This is the worse I could do simulating from 10Hz down to 0.1Hz so to approximate flicker output noise interest region for clocks. I threw in a 1mF filter cap on Vref for good measure. I am not a finite time sample integrator to FFT and average that for ever, :D but can it be 100-120nV p-p roughly?
 

Attachments

  • worse.gif
    worse.gif
    19.8 KB · Views: 443
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.