No, he wasn't. He was talking about a *4 voltage gain that happens when the
feedback resistor ratio is 3:1 in a non-inverting amplifier.
With regard to LEDs, red ones usually are the best (noise voltage normalized to Vf) and
blue ones are the worst. Red ones have lower Vf than blue ones.
Disclaimer: I have not yet tested infrared etc.
regards, Gerhard
That retreat to your personal taste does not come unexpectedly.
best regards, Gerhard
The Silence Of The Amps, what could one desire more?
.
well, thank you John! Now you're talking business. I don't believe in this 'low noise super regulator' nonsense as well.
Please elaborate on line rejection, load rejection, stability requirements etc..
I would love to read your opinions and insights!
regards
No, if you read what I wrote you'll see I was talking about 4 identical devices in series.
And indeed its not absolutely essential for the point. But its fairly poor engineering practice to put a high noise device in series with a low noise one. If you care about noise you won't **** away the advantages of the lower noise (read expensive) device by combining it with a higher noise one. OTOH TL431s are dirt-cheap so if you don't care about noise so much you've got yourself a very cost-effective reference.
Yes, agreed. Orthogonal to the original point.
@gerhard : Red LEDs also had the lowest dynamic impedance, when I tested.
Last edited:
I guess, the fuss is about secret knowledge of filtering and buffered discovered by some genius back in 1050'th. "Learn and grow!" (C)
Gerhard, you still are not being very practical about power supply design. Feedback will NOT reduce the noise of either the reference OR the input stage of the gain amp. In fact, the feedback resistors might INCREASE IT. Negative feedback does not fix everything. That is just an engineering fact, but it can be wonderful for servos and other industrial applications including telephones where it was originally conceived and developed.
For the record, SY is right about stacking devices. The 'reason' it works is that the REFERENCE VOLTAGE is higher, so the reference does not have to be amplified as much, in order to generate the required regulated output voltage.
It reminds me of magnetic tape and adding tape width, like from 1/4 inch to 1/2 inch. It lowers the noise in much the same way. The ACTUAL noise is higher by 3dB, BUT the signal goes up 6dB. See how it works?
Now, what about feedback power supplies in general. There are two distinct problems with regulated power supplies: First, is the noisy reference, like a Zener. This can be improved by RC filtering, BUT the filter is finite, and you often will get ' transient noisebumps' from 1/f noise that is NOT filtered enough.
The second, equally difficult problem is the TRANSIENT PERFORMANCE of the regulator. You see, a feedback regulator is essentially unstable, because it sees a capacitive load. Therefore, when you challenge the regulator with a transient, it will not be able to easily suppress it. JUST LOOK at the data sheets!
So, in conclusion, even IF a version of the LM317 could be developed that was quiet enough, it would still have lousy transient response, and this will effect the sound.
It reminds me of magnetic tape and adding tape width, like from 1/4 inch to 1/2 inch. It lowers the noise in much the same way. The ACTUAL noise is higher by 3dB, BUT the signal goes up 6dB. See how it works?
Now, what about feedback power supplies in general. There are two distinct problems with regulated power supplies: First, is the noisy reference, like a Zener. This can be improved by RC filtering, BUT the filter is finite, and you often will get ' transient noisebumps' from 1/f noise that is NOT filtered enough.
The second, equally difficult problem is the TRANSIENT PERFORMANCE of the regulator. You see, a feedback regulator is essentially unstable, because it sees a capacitive load. Therefore, when you challenge the regulator with a transient, it will not be able to easily suppress it. JUST LOOK at the data sheets!
So, in conclusion, even IF a version of the LM317 could be developed that was quiet enough, it would still have lousy transient response, and this will effect the sound.
Last edited:
I guess, the fuss is about secret knowledge of filtering and buffered discovered by some genius back in 1050'th. "Learn and grow!" (C)
LOL
JC said:
But what kind of horrid transients is a line amp going to present?
uAmps or at worst a few mAmps? I mean if it's good enough to reject high frequency noise well, isn't it good enough to control the output voltage at audio frequencies?
I don't know anything about the LM317, but I'm thinking a shunt regulator that has significantly more current on tap than bias current of a class A preamp, and a short feedback path.
I'll have to build another shunt and see what I can measure at the output of the regulator when presented with some unruly transients.
disclaimer: I have no experience here ...
John, the LM317 spec on load and/or line is not great BUT take a look at the magnitude of those test signals. Huge and unlikely to ever be encountered in a reasonable design.
Secondly, if you run ALL of your circuitry class A, which is not an unreasonable design goal for any low level circuit, you will never see these kinds of problems.
That said, I do agree with the general sentiment that the 317 is not really good enough for top flight audio without some circuit augmentation, some of which has already been proposed by the various participants above.
Secondly, if you run ALL of your circuitry class A, which is not an unreasonable design goal for any low level circuit, you will never see these kinds of problems.
That said, I do agree with the general sentiment that the 317 is not really good enough for top flight audio without some circuit augmentation, some of which has already been proposed by the various participants above.
Sy,
This is true for the reference itself, but not necessarily for a complete circuit.
Let's assume a Shunt regulator made from a pair of transistors (an PNP as Diff Amp and a NPN Darlington or N-Channel Fet power transistor as actual shunt).
The Voltage reference is not buffered or filtered and directly looped into the Emitter line of the PNP. This is the main difference to a Salas Shunt BTW for this style. Tempco will not be given consideration (for a real reg it needs to), only noise.
Let us for starters use a single Red LED at 1.8V Vf, as we want around 10mA into the LED we add a CCS or Resistor from our regulated voltage to get enough current. Reference Noise will be around 400nV based on measurements published here.
The Collector is loaded with a 1mA CCS and drives the power transistor shunt. Let our transistor be a 2SA970 and we see around 100nV noise contribution from the transistor itself. As noises add based on the power we expect around 420nV Noise.
We now use a standard resistive divider to set our output voltage. Let's say we want to regulate +60V for a linestage. We will have around 3uA base current for our PNP Transistor, so let's set the current in the dvider to around 300uA.
Our Reference voltage is 1.8V + 0.6V = 2.4V and our resistor divider is 180K & 7.5K. We bypass the divider with a Capacitor that has a low enough impedance at 20Hz to not increase our noise appreciably (looks like around 47uF will do the job).
Our noise will be pretty much the reference noise, as the noise gain of the circuit will be unity to very low frequencies.
So we get 60V with 420nV of noise or around 3nV|/Hz. In sim's the Impedance loks in the 10's of mOhm, depending on precise implementation and parts choices, with > 100KHz bandwidth.
Now lets connect several led's in series as reference. Say 8pcs. This gives 9dB higher noise from the reference, or around 1,100nV.
We re-adjust the regulator circuit to work with the new 18V reference. What is the result? 9dB more noise.
Hence stacking references has not netted us an improvement in noise, it has only dis-improved the noise.
When used as reference in a competently designed regulator circuit it is invariably the single lowest noise reference combined with a transistor as "amplified zenner" if needed or a change of the circuit to accommodate a lower reference voltage will give the lowest noise, stacking references instead will always give higher noise.
Any pretensions that stacking reference improves SNR are simply hogwash, some strange faith in that "more is better" that has no basis in reality, with real and competently executed circuits.
Ciao T
This is true for the reference itself, but not necessarily for a complete circuit.
Let's assume a Shunt regulator made from a pair of transistors (an PNP as Diff Amp and a NPN Darlington or N-Channel Fet power transistor as actual shunt).
The Voltage reference is not buffered or filtered and directly looped into the Emitter line of the PNP. This is the main difference to a Salas Shunt BTW for this style. Tempco will not be given consideration (for a real reg it needs to), only noise.
Let us for starters use a single Red LED at 1.8V Vf, as we want around 10mA into the LED we add a CCS or Resistor from our regulated voltage to get enough current. Reference Noise will be around 400nV based on measurements published here.
The Collector is loaded with a 1mA CCS and drives the power transistor shunt. Let our transistor be a 2SA970 and we see around 100nV noise contribution from the transistor itself. As noises add based on the power we expect around 420nV Noise.
We now use a standard resistive divider to set our output voltage. Let's say we want to regulate +60V for a linestage. We will have around 3uA base current for our PNP Transistor, so let's set the current in the dvider to around 300uA.
Our Reference voltage is 1.8V + 0.6V = 2.4V and our resistor divider is 180K & 7.5K. We bypass the divider with a Capacitor that has a low enough impedance at 20Hz to not increase our noise appreciably (looks like around 47uF will do the job).
Our noise will be pretty much the reference noise, as the noise gain of the circuit will be unity to very low frequencies.
So we get 60V with 420nV of noise or around 3nV|/Hz. In sim's the Impedance loks in the 10's of mOhm, depending on precise implementation and parts choices, with > 100KHz bandwidth.
Now lets connect several led's in series as reference. Say 8pcs. This gives 9dB higher noise from the reference, or around 1,100nV.
We re-adjust the regulator circuit to work with the new 18V reference. What is the result? 9dB more noise.
Hence stacking references has not netted us an improvement in noise, it has only dis-improved the noise.
When used as reference in a competently designed regulator circuit it is invariably the single lowest noise reference combined with a transistor as "amplified zenner" if needed or a change of the circuit to accommodate a lower reference voltage will give the lowest noise, stacking references instead will always give higher noise.
Any pretensions that stacking reference improves SNR are simply hogwash, some strange faith in that "more is better" that has no basis in reality, with real and competently executed circuits.
Ciao T
Hi,
First, resistor tolerances that are better than those of the TL431 are readily available at a cost of fractions of a cent per pcs. So the "improved accuracy" argument is highly specious.
Secondly, bypassing the divider that sets the voltage is mandatory for low noise. Due to the impedances involved there is no need for excotic low distortion types.
In this case the competently designed circuit trumps both of those you suggest.
We may even go one further.
Let us say we want a 60V Reference (to for example supply a preamp with a emitter follower or source follower), we could use a TL431 and a pair of NPN Transistor with suitable resistors (180K + 10K//39K) and a 47uF bypass capacitor to get the same audio band noise as the TL431 alone.
I repeat, "stacking" references is a very stupid way to get low noise, as it does not get low noise and only unnecessarily increases component count. I can see how the "more is better" concept appeals to those who place faith into certain audiophile notions, but evidence argues strongly against such a practice.
Faith is no substitute for sound (pun intended) engineering.
BTW, see my previous example for a much lower noise 60V source...
Ciao T
First, resistor tolerances that are better than those of the TL431 are readily available at a cost of fractions of a cent per pcs. So the "improved accuracy" argument is highly specious.
Secondly, bypassing the divider that sets the voltage is mandatory for low noise. Due to the impedances involved there is no need for excotic low distortion types.
In this case the competently designed circuit trumps both of those you suggest.
We may even go one further.
Let us say we want a 60V Reference (to for example supply a preamp with a emitter follower or source follower), we could use a TL431 and a pair of NPN Transistor with suitable resistors (180K + 10K//39K) and a 47uF bypass capacitor to get the same audio band noise as the TL431 alone.
I repeat, "stacking" references is a very stupid way to get low noise, as it does not get low noise and only unnecessarily increases component count. I can see how the "more is better" concept appeals to those who place faith into certain audiophile notions, but evidence argues strongly against such a practice.
Faith is no substitute for sound (pun intended) engineering.
BTW, see my previous example for a much lower noise 60V source...
Ciao T
Sy,
Folded cascodes (such as I use in the -147dBVA Ein - AMR product that I shall not mention to avoid being chided by you for "promoting it relentlessly") have excellent PSRR, competent implementation given.
Conventional cascodes have excellent PSRR if we reference the output to the supply rail and not to "ground", again, this falls under competent implementation or as it is so cutely called "skilled in the art".
I wish you would avoid constantly posting sweeping generalisations that are so obviously untrue, they make me spit out my morning coffee...
It would save me time cleaning my laptop and having to refute them...
Ciao T
Folded cascodes (such as I use in the -147dBVA Ein - AMR product that I shall not mention to avoid being chided by you for "promoting it relentlessly") have excellent PSRR, competent implementation given.
Conventional cascodes have excellent PSRR if we reference the output to the supply rail and not to "ground", again, this falls under competent implementation or as it is so cutely called "skilled in the art".
I wish you would avoid constantly posting sweeping generalisations that are so obviously untrue, they make me spit out my morning coffee...
It would save me time cleaning my laptop and having to refute them...
Ciao T
- Status
- Not open for further replies.