Christer
If you hunt for previous posts from me you may find just the noise comparison you desire - if I get a moment, I'll look myself and post a link.
EDIT:
http://www.diyaudio.com/forums/showthread.php?threadid=985&perpage=15&pagenumber=2
Bottom of the page for a plot - those TL431's really are crap...
P.S. there's not too much difference in noise between LT and LM 317's, but impedance is a bit lower, generally, for the LT parts.
If you hunt for previous posts from me you may find just the noise comparison you desire - if I get a moment, I'll look myself and post a link.
EDIT:
http://www.diyaudio.com/forums/showthread.php?threadid=985&perpage=15&pagenumber=2
Bottom of the page for a plot - those TL431's really are crap...
P.S. there's not too much difference in noise between LT and LM 317's, but impedance is a bit lower, generally, for the LT parts.
Re: Regulators
Elso,
I think settling time and transient performance is important
for DAC regs. Jung regs would have poor transient performance
due to their super high open loop gain required to acheive such
low OP Z (in audio band). Above the audio band the Jung reg
falls over WRT OP Z.
Just use a simple discrete shunt or follower and relaxation
is restored.
Terry
Elso Kwak said:
Elso,
I think settling time and transient performance is important
for DAC regs. Jung regs would have poor transient performance
due to their super high open loop gain required to acheive such
low OP Z (in audio band). Above the audio band the Jung reg
falls over WRT OP Z.
Just use a simple discrete shunt or follower and relaxation
is restored.
Terry
Why etc
Hi Terry,
Having an obvious interest in the Jung-type regulators, I find your statements interesting. If I use a simple discrete shunt or follower, the DAC hi-speed load currents cause lots of hi-freq spikes and noise on the supply lines that is larger (much) than with the Jung regulators. This is verifiable. Yet, according to you, this would improve the sonics. Do you have any explanation to this effect?
Jan Didden
Hi Terry,
Having an obvious interest in the Jung-type regulators, I find your statements interesting. If I use a simple discrete shunt or follower, the DAC hi-speed load currents cause lots of hi-freq spikes and noise on the supply lines that is larger (much) than with the Jung regulators. This is verifiable. Yet, according to you, this would improve the sonics. Do you have any explanation to this effect?
Jan Didden
Re: Christer
I think you may have misunderstood that noone has said that a 431 is a low noise device. If you want to have low noise you must first filter the zener voltage. There is no way around it. Elso does this with a big cap. Remember also that a zener is rather low ohmic (dynamic impedance) which demands for a rather large cap.
I want also to point out that the noise floor of the measurement equipment isn't presented, nor the measurement itself. I suspect that noise is picked up also. I haven't so great experience of FFT but it's very easy to make errors in these measurements.
ALW said:
I think you may have misunderstood that noone has said that a 431 is a low noise device. If you want to have low noise you must first filter the zener voltage. There is no way around it. Elso does this with a big cap. Remember also that a zener is rather low ohmic (dynamic impedance) which demands for a rather large cap.
I want also to point out that the noise floor of the measurement equipment isn't presented, nor the measurement itself. I suspect that noise is picked up also. I haven't so great experience of FFT but it's very easy to make errors in these measurements.
Re: Christer
OK, thanks. I vaguely remember that thread now. Anyway, your
comparison is interesting, but it doesn't really say that much if
you only tested two LT1086s and one each of the others. The
difference in noise between 317 and 1086 is somewhere around
5dB which could easily be explained by the statistical
variations. It is interesting, though, that the noise spectra of
317 and 1086 are almost identical, except for the almost
constant difference in level.
It would be very interesting if someone who uses a lot of both
types of regulators could do a similar analysis with a larger
number of units of each type.
ALW said:
OK, thanks. I vaguely remember that thread now. Anyway, your
comparison is interesting, but it doesn't really say that much if
you only tested two LT1086s and one each of the others. The
difference in noise between 317 and 1086 is somewhere around
5dB which could easily be explained by the statistical
variations. It is interesting, though, that the noise spectra of
317 and 1086 are almost identical, except for the almost
constant difference in level.
It would be very interesting if someone who uses a lot of both
types of regulators could do a similar analysis with a larger
number of units of each type.
Quickies
Re: TL431
Peranders,
Some people actually rate TL431 as the be-all and end-all of PSU's, not on this well-informed place though.
They are poor in both noise and impedance, when used alone.
You are right though - filter them and they make good references.
Noise floor of the measurement system, in this instance, would be about 10-15 dB below the bottom of the graticule. My current x100 differential buffer amp arrangement lowers this much further, as the Jung reg's dropped significantly below that noise floor The current system can resolve PSU noise around 3nV / rt Hz, with careful screening.
I'm not sure what it was for those measurements - I was less fastidious in absolute measurements then, as ratiometric comparisons were my intention for simplicity - lower = better for noise and impedance.
Re: Sample qty.
Christer,
The FFT analyser can only store 4 traces, suffice to say that in some >100 units of LM317 and around 50 off 1086's used they show no statistically significant variation in noise performance (1 dB max).
The LM317AT shows improved impedance performance for those that wish to try it.
Re: TL431
Peranders,
Some people actually rate TL431 as the be-all and end-all of PSU's, not on this well-informed place though.
They are poor in both noise and impedance, when used alone.
You are right though - filter them and they make good references.
Noise floor of the measurement system, in this instance, would be about 10-15 dB below the bottom of the graticule. My current x100 differential buffer amp arrangement lowers this much further, as the Jung reg's dropped significantly below that noise floor The current system can resolve PSU noise around 3nV / rt Hz, with careful screening.
I'm not sure what it was for those measurements - I was less fastidious in absolute measurements then, as ratiometric comparisons were my intention for simplicity - lower = better for noise and impedance.
Re: Sample qty.
Christer,
The FFT analyser can only store 4 traces, suffice to say that in some >100 units of LM317 and around 50 off 1086's used they show no statistically significant variation in noise performance (1 dB max).
The LM317AT shows improved impedance performance for those that wish to try it.
Re: Quickies
OK, I might have misunderstood you. This makes it more
interesting. A sample size of 50 should give reasonable
statistical significance, at least if they are not all from the same
batch.
ALW said:
OK, I might have misunderstood you. This makes it more
interesting. A sample size of 50 should give reasonable
statistical significance, at least if they are not all from the same
batch.
Re: Why etc
The OP Z of the jung reg shoots up at only a few hundred
kHz. The OP Z of a good discrete can be kept quite low AND
flat to 10's of Mhz before it goes south.
But the result was attained from listening by me and others.
Flatness of OP Z and settling time was the only thing I could
corelate to naturalness of sound. They all have very low noise.
Perhaps a better measurement would be wide band
FFT at DAC analog OP?
Maybe you should ask yourself why everyone here is
so interested in replacing their OPA based I-V's with
resistors, transformers and 0 FB cascodes when
the OPA I-V's can be shown to measure superior!
It's not always cut and dry on a scope or APII.
Terry
janneman said:
The OP Z of the jung reg shoots up at only a few hundred
kHz. The OP Z of a good discrete can be kept quite low AND
flat to 10's of Mhz before it goes south.
But the result was attained from listening by me and others.
Flatness of OP Z and settling time was the only thing I could
corelate to naturalness of sound. They all have very low noise.
Perhaps a better measurement would be wide band
FFT at DAC analog OP?
Maybe you should ask yourself why everyone here is
so interested in replacing their OPA based I-V's with
resistors, transformers and 0 FB cascodes when
the OPA I-V's can be shown to measure superior!
It's not always cut and dry on a scope or APII.
Terry
OP Z
That may be slightly colourful language, but whilst the Jung reg op Z does rise generally with increasing frequency, suitable choice of error amp, coupled with high performance o/p decoupling can minize any op Z humps at the expense of ultimate op Z performance if desired.
This is one of the reason I no longer use the AD797, except for clock supplies, where it's noise figure is hugely beneficial, as there does appear to be sonic merit, in analogue supplies, to maintaining a more constant op Z.
It's also the reason I like the POOGE discrete reg - for its simplicity, it achieves a ruler flat o/p Z over a wide bandwidth. Noise figure at LF is worse than 3-terminal devices, but it sounds WAY better than any of them.
A more complicated discrete design can do this better, but I very strongly doubt that any discrete reg. that maintains this over 10's of MHz actively has particularly low o/p Z to start with, which leaves me doubtful as to whether topology, or implementation, is the issue with those that prefer 'worse' PSU's. When I find such results myself, I tend to investigate further, and have many times found problems that lay elsewhere, but were masked by the 'worse' supply.
At the end of the day though, if it sounds better, it is better, I just like to know why - I'm like that 😉
Andy.
That may be slightly colourful language, but whilst the Jung reg op Z does rise generally with increasing frequency, suitable choice of error amp, coupled with high performance o/p decoupling can minize any op Z humps at the expense of ultimate op Z performance if desired.
This is one of the reason I no longer use the AD797, except for clock supplies, where it's noise figure is hugely beneficial, as there does appear to be sonic merit, in analogue supplies, to maintaining a more constant op Z.
It's also the reason I like the POOGE discrete reg - for its simplicity, it achieves a ruler flat o/p Z over a wide bandwidth. Noise figure at LF is worse than 3-terminal devices, but it sounds WAY better than any of them.
A more complicated discrete design can do this better, but I very strongly doubt that any discrete reg. that maintains this over 10's of MHz actively has particularly low o/p Z to start with, which leaves me doubtful as to whether topology, or implementation, is the issue with those that prefer 'worse' PSU's. When I find such results myself, I tend to investigate further, and have many times found problems that lay elsewhere, but were masked by the 'worse' supply.
At the end of the day though, if it sounds better, it is better, I just like to know why - I'm like that 😉
Andy.
hi alw,
could you please post the scematics for the +/- POOGE discrete reg ?
best regards,
troels
could you please post the scematics for the +/- POOGE discrete reg ?
best regards,
troels
Re: Re: Why etc
Terry, it is true that the Jung regs shoot up, as you say. But if you compare it with the discretes, it actually is the other way around. The Jung "shoots down", and is generally much lower than discretes below supersonic frequencies. If you can have extra low Z at some part of the range (the important midrange!), I fail to see that flatness in itself is a benefit.
I readily accept that you heard a difference, but what was it? Just different, worse, better? We too often assume that a difference MUST be an improvement. Having made a change anticipating and actually actively listening for an improvement, it is very, very hard to find anything to the contrary.
And as for why everyone wants to change to transformers & open loop stuff, well, for one it is much easier to do, you need much less technical expertise, you don't need sophisticated test equipment to see if all is well. It has an elegant attraction to it: put in one of these nice transformers, weighting at leat two pounds, now there is quality! See, I put it in, and how much the sound field opened up! Incredible!
But I am sure there are other reasons as well.
Jan Didden
Terry Demol said:
Terry, it is true that the Jung regs shoot up, as you say. But if you compare it with the discretes, it actually is the other way around. The Jung "shoots down", and is generally much lower than discretes below supersonic frequencies. If you can have extra low Z at some part of the range (the important midrange!), I fail to see that flatness in itself is a benefit.
I readily accept that you heard a difference, but what was it? Just different, worse, better? We too often assume that a difference MUST be an improvement. Having made a change anticipating and actually actively listening for an improvement, it is very, very hard to find anything to the contrary.
And as for why everyone wants to change to transformers & open loop stuff, well, for one it is much easier to do, you need much less technical expertise, you don't need sophisticated test equipment to see if all is well. It has an elegant attraction to it: put in one of these nice transformers, weighting at leat two pounds, now there is quality! See, I put it in, and how much the sound field opened up! Incredible!
But I am sure there are other reasons as well.
Jan Didden
Where's Jocko when we need him ?
Anyhow, please see this thread: http://www.diyaudio.com/forums/showthread.php?s=&threadid=6121
OK, I know better but I still use the OPA's mostly because of convenience... Have experience with discrete built I/V though and they can be a lot better than even the best opamps when designed right. But results till now were very good, only voluminous and difficult to build in the cdplayer. I am about to make a replacement board for a cdplayer. It will contain a discrete built I/V.
Jan, the tone of your posts gives me the idea that one is stupid not to believe the standard stuff and that it is a waste of time wanting to try transformers or such things. And open loop designs are nonsense too according to you because one doesn't need technical experience. Come on man, wake up !
Discrete POOGE regulator
http://www.diyaudio.com/forums/showthread.php?s=&threadid=5336&highlight=pooge+AND+discrete
😉
Hi Troels, You mean this one from 1995:tbla said:hi alw,
could you please post the scematics for the +/- POOGE discrete reg ?
best regards,
troels
http://www.diyaudio.com/forums/showthread.php?s=&threadid=5336&highlight=pooge+AND+discrete
😉
Re: Quickies
431 is the cheapest "variable zener" on the market and also very good, to be a zener. The dynamic impedance is very low but it is still a zener-like component. 431 is at it's best if it is used as a voltage reference.
ALW said:
431 is the cheapest "variable zener" on the market and also very good, to be a zener. The dynamic impedance is very low but it is still a zener-like component. 431 is at it's best if it is used as a voltage reference.
Further Improved Poer supply
Hi All,
Attached is a further improved powersupply circuit.
The red LED and the PNP transistor provide a 2mA constant current source for the LM329. The red LED is used as a 1.6V reference providing a 1V drop across the 470 Ohm resistor in the emittor lead.
This circuit also works with four green LED's, in a series string, in place of the LM329 but sound is better with the circuit as depicted.
I also did some experiments with LT1021-5 and MAX 6250. These are expensive low-noise 5V references. I amplified the 5V to 6.9V with a OP27 and connected this to the base of Q1. Sound is definitely different but I am not sure it is better. It has a kind of inflated quality just as the sound of commercials on our national television.😕 Also more fatigueing.😕 I keep lowering the volume, not a good sign...😱
Hi All,
Attached is a further improved powersupply circuit.
The red LED and the PNP transistor provide a 2mA constant current source for the LM329. The red LED is used as a 1.6V reference providing a 1V drop across the 470 Ohm resistor in the emittor lead.
This circuit also works with four green LED's, in a series string, in place of the LM329 but sound is better with the circuit as depicted.
I also did some experiments with LT1021-5 and MAX 6250. These are expensive low-noise 5V references. I amplified the 5V to 6.9V with a OP27 and connected this to the base of Q1. Sound is definitely different but I am not sure it is better. It has a kind of inflated quality just as the sound of commercials on our national television.😕 Also more fatigueing.😕 I keep lowering the volume, not a good sign...😱
Attachments
Correction
Hi,
I forgot the two 4700µF caps in parallel with C10 to include in the picture. Of course you can also use a higher value of choke and less capacitance. 😱 😉
Hi,
I forgot the two 4700µF caps in parallel with C10 to include in the picture. Of course you can also use a higher value of choke and less capacitance. 😱 😉
Hi Elso,
What is the reason you are still using a “triplet”? The output impedance is mainly defined by the current flowing out of the emitter of Q3. For this reason the circuit should have a minimum standing load current of over 10mA to 25mA. So it is wise to add a load resistance of 220 ohms to the output. Or does the supply modulation due to the high output impedance + varying load current add to the sonic qualities of the circuit? And why are you loading the driving transistors to ground? (R1 and R3). A resistor between the base and the emitter is a more efficient way. I don’s see any advantages over a normal darlington build with a BC649 and a BD135 (See my pervious posted circuit). Your circuit is still spitting out the excessive noise of the reference zener directly. Take rid of it at the origin: the zener itself and not by large output caps.
Feeding the zener by a current source only reduces ripple, but not noise
😕
What is the reason you are still using a “triplet”? The output impedance is mainly defined by the current flowing out of the emitter of Q3. For this reason the circuit should have a minimum standing load current of over 10mA to 25mA. So it is wise to add a load resistance of 220 ohms to the output. Or does the supply modulation due to the high output impedance + varying load current add to the sonic qualities of the circuit? And why are you loading the driving transistors to ground? (R1 and R3). A resistor between the base and the emitter is a more efficient way. I don’s see any advantages over a normal darlington build with a BC649 and a BD135 (See my pervious posted circuit). Your circuit is still spitting out the excessive noise of the reference zener directly. Take rid of it at the origin: the zener itself and not by large output caps.
Feeding the zener by a current source only reduces ripple, but not noise
😕
According to some sources /LC Audio/ the noise in the power rail can seriously damage the sound due to inducting jitter in the clock generator. You, Patwen, can check this out by a connecting a passive RC filter right in front of the Clock generator IC /and before that check which it is/ os, say, 10-100ohms and 1000-2200uF, depending on consumption. Please reply to me if you find something true in this! 10-x
- Status
- Not open for further replies.