Q: Did anybody check the claim that 2X2=4?
A: Yes, I did on my toes.
Don't try a 4x4 on your toes.
regards, Gerhard
I want a t-shirt with your .sig
An externally hosted image should be here but it was not working when we last tested it.
regards, Gerhard
I want a t-shirt with your .sig
We'll have an Exchange Student from Germany next month in our house. May be I will print one for you.
Jan,
The problem with selling your services is that you want to publish just enough to let folks know that you ain't stoopid but not enough that they can do it themselves. The Wenzel paper has been around a long time and taught a lot of folks. Some of whom actually realize the limits and improve on the circuit. You know folks like you & T.
I assume that there are bypasses in the oscillator circuits fed from his lower noise supply so that at his frequencies of interest the power rail impedance is close to zero. As has been mentioned (and references posted by you!) paralleling capacitors to clean up a supply rail has potential problems.
In reality the parallel capacitor simulations often shown to prove that point do not have good capacitor models. Often this means that the capacitor model is "improved" by adding inductance thus making the problem clearer. However in real capacitors there are also interesting losses some of which are non-linear. These are rarely modeled and if modeled are just lumped into a single resistor.
As to an op-amp version it would be difficult to get a working circuit at the frequencies of interest to a crystal oscillator!
ES
John,
In order to teach anyone that person must be both willing to audit the lesson and capable to absorb it's content.
However, as the subject may be of some interest in more general terms.
In this case, let's design a powersupply that uses LM317 and is as optimised as it gets. If that is good enough depends on many factors.
Let us in this case state that we will supply an audio circuit that will draw 100mA or less and with +/-12.5V.
For starters I would use a pair of independent secondaries on the mains transformer, with individual bridge rectifiers using schottky diodes. I would then use a CLC filter with inductors in both + & - line.
Selection of the inductor depends on many factors, it may be necessary to connect damping resistors across the capacitors and inductors to limit resonant behaviour at very low frequencies. Big value capacitors help, some DCR in the chokes can help manage resonances as well.
I am normally aiming so that the 100Hz is filtered enough that with the ripple rejection of the regulator pushed the ripple below the regulators self noise (around 25uV RMS or -92dBV for 20KHz bandwidth).
I would then use the LM317 for both rails (not 317/337) and use resistor values that will draw around 100mA current, to wit, 12 Ohm and 110 Ohm to set the DC voltage will give 12.7V design centre.
The 110 Ohm resistor must bypassed to to set the AC gain to unity, if we just scale the datasheet values we may be tempted to use 100uF. However a 3,300uF capacitor will make sure than only below around 6Hz do noise and output impedance rise.
The LM317 output impedance scales with current, in our case we expect more than 100 and less than 200mA, so for a design centre of 150mA total current we get around 7mOhm output impedance from the LM317. This will double at around 1.2KHz.
If we wish to offset this inductive behaviour of the regulator (which is really just insufficient loop gain at higher frequencies), we need a cap is around 7mOhm ESR and around 7mOhm Xc at 1.2KHz. a 2,200uF capacitor will answer well.
However so-called "low impedance" electrolytics will tend to have around 20 -30 mOhm ESR, if we can tolerate this level of output impedance above around 2KHz, we can use a single 3,300uF cap on the output of the regulator.
The regulator systems output impedance will be flat from around 10Hz to 1KHz at around 7mOhm rising to 20...30mOhm at around 4KHz and flat to where the cap's become inductive. Transient load behaviour basically reflects the capacitors ESR and ESL, also higher frequency noise is progressively attenuated.
If we postulate a phono stage with OPA637 running at 40dB gain (like my old "Analog Addicts" design) we will have around 45dB PSRR on the worse (positive) rail at 1KHz.
So the output noise induced from the LM317 Supply would be around -137dBV. The OPA637's self noise is around -82dBV, so we are quite a lot below the circuits self noise.
However, to illustrate an example for a case where the noise is too high, the MC pre-pre of the Analog Addicts Pre uses a circuit that has zip PSRR. Design centre output voltage is 5mV or -46dBV.
If we did not use additional filtering of the PSU line, the SNR would only be -92dBV - (-46dBV) = 46dB which is crummy.
Adding 60dB of noise filtering improves the situation to around 106dB noise from the supply lines and below the theoretical noise floor of the actual circuit.
Anyway, this slightly improved LM317 is not the answer to all questions, but it surprisingly good.
Ciao T
In order to teach anyone that person must be both willing to audit the lesson and capable to absorb it's content.
However, as the subject may be of some interest in more general terms.
In this case, let's design a powersupply that uses LM317 and is as optimised as it gets. If that is good enough depends on many factors.
Let us in this case state that we will supply an audio circuit that will draw 100mA or less and with +/-12.5V.
For starters I would use a pair of independent secondaries on the mains transformer, with individual bridge rectifiers using schottky diodes. I would then use a CLC filter with inductors in both + & - line.
Selection of the inductor depends on many factors, it may be necessary to connect damping resistors across the capacitors and inductors to limit resonant behaviour at very low frequencies. Big value capacitors help, some DCR in the chokes can help manage resonances as well.
I am normally aiming so that the 100Hz is filtered enough that with the ripple rejection of the regulator pushed the ripple below the regulators self noise (around 25uV RMS or -92dBV for 20KHz bandwidth).
I would then use the LM317 for both rails (not 317/337) and use resistor values that will draw around 100mA current, to wit, 12 Ohm and 110 Ohm to set the DC voltage will give 12.7V design centre.
The 110 Ohm resistor must bypassed to to set the AC gain to unity, if we just scale the datasheet values we may be tempted to use 100uF. However a 3,300uF capacitor will make sure than only below around 6Hz do noise and output impedance rise.
The LM317 output impedance scales with current, in our case we expect more than 100 and less than 200mA, so for a design centre of 150mA total current we get around 7mOhm output impedance from the LM317. This will double at around 1.2KHz.
If we wish to offset this inductive behaviour of the regulator (which is really just insufficient loop gain at higher frequencies), we need a cap is around 7mOhm ESR and around 7mOhm Xc at 1.2KHz. a 2,200uF capacitor will answer well.
However so-called "low impedance" electrolytics will tend to have around 20 -30 mOhm ESR, if we can tolerate this level of output impedance above around 2KHz, we can use a single 3,300uF cap on the output of the regulator.
The regulator systems output impedance will be flat from around 10Hz to 1KHz at around 7mOhm rising to 20...30mOhm at around 4KHz and flat to where the cap's become inductive. Transient load behaviour basically reflects the capacitors ESR and ESL, also higher frequency noise is progressively attenuated.
If we postulate a phono stage with OPA637 running at 40dB gain (like my old "Analog Addicts" design) we will have around 45dB PSRR on the worse (positive) rail at 1KHz.
So the output noise induced from the LM317 Supply would be around -137dBV. The OPA637's self noise is around -82dBV, so we are quite a lot below the circuits self noise.
However, to illustrate an example for a case where the noise is too high, the MC pre-pre of the Analog Addicts Pre uses a circuit that has zip PSRR. Design centre output voltage is 5mV or -46dBV.
If we did not use additional filtering of the PSU line, the SNR would only be -92dBV - (-46dBV) = 46dB which is crummy.
Adding 60dB of noise filtering improves the situation to around 106dB noise from the supply lines and below the theoretical noise floor of the actual circuit.
Anyway, this slightly improved LM317 is not the answer to all questions, but it surprisingly good.
Ciao T
Good point. I think (but have not tried) that with 0.1% resistors and a fast ((A)DSL driver) opamp I can get up to 60dB reduction at maybe up to 50MHz when pushed.
Edit: this can be easily scaled to a series R of say 5 ohms instead of 15 to improve Zout and limit load-current-induced riple.
jan
Early IR3601 results
Back to low noise stuff. I have some prelim measurements on the IF3601’s (BIG N channel JFETs) that were a kind Christmas gift from Pass Labs. Curve fitting Vgs and Id at a few spaced points gave me about -.83V Vp and a beta (transconductance parameter) of .2, this is BIG. That gives the Idss at 138mA and gm @ 10mA of 89mS. I made a little common source amplifier with 200 Ohm drain load at 10mA drain current and measured about .4nV spot noise and low 1/f corner frequency comparable to 2SK170. This is compatible with the 89mS but not with the stated .3nV at 5mA drain current, in fact the gm at Idss is 330mS about half of that stated on the data sheet. Cin is big, this would be MC only.
Still at 50mA drain current I figure with one FET you could get .25nV spot noise. I wish the case (TO99) was not the gate so heat sinking (if needed) would not require some care.
BTW I was surprised at the return of the RTI noise discussion. I walked a friend through the same thing with the Earthworks pre-amp. They spec .6nV spot noise and -143dbVA, .6nV is 85nVrms 20-20k which is -141.4dbV and for a flat spectrum A weighting gives back ~1.8dB for -143.2dBVA, close enough. For .4nV (the Blowtorch number) it’s 57nVrms 20-20k and -144.9dbV or -146.7dBVA, again close enough. Unweighted noise has another issue in that it would need a 20-20k brickwall filter or the filter skirts would have to be factored out, the A weighting essentially eliminates these effects.
Back to low noise stuff. I have some prelim measurements on the IF3601’s (BIG N channel JFETs) that were a kind Christmas gift from Pass Labs. Curve fitting Vgs and Id at a few spaced points gave me about -.83V Vp and a beta (transconductance parameter) of .2, this is BIG. That gives the Idss at 138mA and gm @ 10mA of 89mS. I made a little common source amplifier with 200 Ohm drain load at 10mA drain current and measured about .4nV spot noise and low 1/f corner frequency comparable to 2SK170. This is compatible with the 89mS but not with the stated .3nV at 5mA drain current, in fact the gm at Idss is 330mS about half of that stated on the data sheet. Cin is big, this would be MC only.
Still at 50mA drain current I figure with one FET you could get .25nV spot noise. I wish the case (TO99) was not the gate so heat sinking (if needed) would not require some care.
BTW I was surprised at the return of the RTI noise discussion. I walked a friend through the same thing with the Earthworks pre-amp. They spec .6nV spot noise and -143dbVA, .6nV is 85nVrms 20-20k which is -141.4dbV and for a flat spectrum A weighting gives back ~1.8dB for -143.2dBVA, close enough. For .4nV (the Blowtorch number) it’s 57nVrms 20-20k and -144.9dbV or -146.7dBVA, again close enough. Unweighted noise has another issue in that it would need a 20-20k brickwall filter or the filter skirts would have to be factored out, the A weighting essentially eliminates these effects.
Don't try a 4x4 on your toes.
regards, Gerhard
I want a t-shirt with your .sig
I heard in WV they can do 3+3 on one foot!
What about the use of an external reference such the LM 329 for a 3 terminal regulator. See high stability regulator on page 7 of Lt 1033 datasheet.
http://cds.linear.com/docs/Datasheet/1033fc.pdf
Still needs oversized bypass caps and preloading and raw supply as worked out by Thorsten.
http://cds.linear.com/docs/Datasheet/1033fc.pdf
Still needs oversized bypass caps and preloading and raw supply as worked out by Thorsten.
Was your frequency low enough when you determined the transconductance from
the cs amplifier?
Last week I tested an Avago PHEMT as a 5V 50 Ohm digital driver (CS, current source from 9V,
Schottky clamp to 5V, force that node down to 0V with the phemt if low level
is wanted) and got a disappointing 800 ps rise/fall time. Turned out, it was the Miller
slowing down the 50 Ohm source of the Anritsu pulse pattern generator. The gate
voltage was just as slow. Perfect integrator
-- and I love these Agilent active probes!With the 200 pF feedback capacitance you could face a similar effect.
Yesterday, I got the SOT-89 version of the PHEMTs, they can do 1A, that should make
one fast sounding headphone amplifier...
With JFETs, the substrate is usually gate, too, so that the channel can be squeezed
from top and bottom at the same time. So, the gate is on that pin that is the home
of the collector on bipolars.
\<Gerhard>
Last edited:
The gm was from a DC plot. I refined my measurements and got .35nV at 10mA which is closer and an exact fit to .089S. BTW I measured the Cgd at 5V to be ~110pF.
Last edited:
I think it's an interesting approach, but choke winding capacitance and cap ESR and ESL are still issues and it's bulky and expensive. For opamp work a 317 with heavy RC filtering at the opamp pins gets you most of the way. If you want a step up from that, add a ripple eater (aka gyrator). I see John uses 317+ripple eater+finesser which clearly delivers results for him on his discrete based circuits.
I can't comment on what a DIYer without a scope and decent tools should do. I am very lucky to not have that issue (except for an AP). If an inductor and a cap does it for you then that's all that's important.
Last edited:
Hi,
The Cap's ESR is what it is. At around 30mOhm for the Cap in my Analog Addicts Phono that allows for a lot of attenuation...
The choke will indeed have a winding capacitance, so it will not continue to filter to infinity, however the resonance is typically around 1KHz, where the impedance can be very high.
With the particular choke and capacitor used in the AA Preamp we get, accounting for parasitics etc. > 70dB noise rejection at 100Hz and around 100dB for 1KHz-10KHz. And self-noise is very low.
You need a lot of electronic circuitry to get this level of filtering of regulator noise and you are unlikely to get equally low noise as active parts are way noisier.
As for expensive and bulky, back when I designed the AA Phono (in around '95) these chokes where available from a National Electronics Store Chain in the UK (Maplin) for 3.99pound a piece. Bulky, no worse than the 12V SLC Batteries that where intended to power the rest of the circuit.
But the advantage of choke and Cap was it took only two parts to get low noise, easily hard wired etc... To many people this makes a lot of difference in actually attempting and succeeding at such a project.
You may find that I published an update of the AA in 2007 that used this approach for both main supplies and the MC Pre-Pre. Past that, you will never get as low noise and as low impedance using a ripple eater as you get with RC or LC filtering.
A scope alone does not help for very low noise work.
As for the rest, many of the designs I publish are very simple and aimed at being reasonably easy to replicate by a diy'er with minimal instrumentation and only modest experience.
This means if I can use a simple RC filter or LC filter instead of a complex electronic circuit and get the same or better results I will used and recommend the simple filter circuit over the electronic designs.
What I use in commercial designs of even my own at home varies, though I would note that German Studio gear I worked with and serviced in the 80's extensively used LC and RC filtering and achieved excellent noise performance, so I tend to be partial to them.
What you do not get, apparently, is that the Analogue Addicts Phono was not designed for my own use. I was using an Artur Loesch derived Tube Phono at the time.
At the time members of an e-mail list (the Analogue Addicts) which was primarily England (not UK) based complained about the paucity of high quality and high performance Phono Stages in the sub 500 pound category and the fact that if there was a Phono Stage at all in your Amp or Pre, it was usually quite bad.
At the same time Audio DIY in the UK was going quite strong, in part at least helped by the original Maplin Shop chain.
Maplin carried a most amazing catalogue of electronic parts, including even OPA627 and 637 in metal cans and Philips Polystrene and Tinfoil capacitors, many excellent Audax Speaker Drivers (including the TW34 so beloved by the BBC) crossover components, the whole range of Velleman Kits including the much lusted after K4000 Tube Amp and their own range of Tube Kits (Amp and Preamp) supported by very high quality transformers and chokes also sold separate.
I remember, the set of Iron (Mains TX and Outputs) for a Stereo EL34 Amp (e.g. Dynaco ST-70 copy) could be bought for under 100 Pound.
Anyway, in this particular situation people that had easy access to parts, some modest experience messing with electronics and a great desire to build a high performance Phono.
So I decided to offer an older design of mine, adapted to use parts available from Maplin (and Cricklewood Electronics for the 2SK147) and keeping things very simple, to ensure a high success rate when building the design with limited tools and on a vero-board.
The choice of the choke and of batteries as power source was part of this. For people who did not want to use batteries the same chokes where originally recommended to be put into the main PSU Lines following a simple Vellemann 317/337 based PSU KIT.
If you where not a member of the Analogue Addicts E-Mail list in the mid 90's and did not live in England, the Analougue Addicts Phono was not designed for you.
It was most emphatically not designed for easy and low cost implementation by chinese Kopy-Mao's to turn a profit or to please some technocrat who will complain that it could have all been done at halve price of the choke with around 100 cheap parts using some kind of super hyper mega regulator, which in actual terms would still have had higher noise anyway ...
Ciao T
The Cap's ESR is what it is. At around 30mOhm for the Cap in my Analog Addicts Phono that allows for a lot of attenuation...
The choke will indeed have a winding capacitance, so it will not continue to filter to infinity, however the resonance is typically around 1KHz, where the impedance can be very high.
With the particular choke and capacitor used in the AA Preamp we get, accounting for parasitics etc. > 70dB noise rejection at 100Hz and around 100dB for 1KHz-10KHz. And self-noise is very low.
You need a lot of electronic circuitry to get this level of filtering of regulator noise and you are unlikely to get equally low noise as active parts are way noisier.
As for expensive and bulky, back when I designed the AA Phono (in around '95) these chokes where available from a National Electronics Store Chain in the UK (Maplin) for 3.99pound a piece. Bulky, no worse than the 12V SLC Batteries that where intended to power the rest of the circuit.
But the advantage of choke and Cap was it took only two parts to get low noise, easily hard wired etc... To many people this makes a lot of difference in actually attempting and succeeding at such a project.
You may find that I published an update of the AA in 2007 that used this approach for both main supplies and the MC Pre-Pre. Past that, you will never get as low noise and as low impedance using a ripple eater as you get with RC or LC filtering.
A scope alone does not help for very low noise work.
As for the rest, many of the designs I publish are very simple and aimed at being reasonably easy to replicate by a diy'er with minimal instrumentation and only modest experience.
This means if I can use a simple RC filter or LC filter instead of a complex electronic circuit and get the same or better results I will used and recommend the simple filter circuit over the electronic designs.
What I use in commercial designs of even my own at home varies, though I would note that German Studio gear I worked with and serviced in the 80's extensively used LC and RC filtering and achieved excellent noise performance, so I tend to be partial to them.
What you do not get, apparently, is that the Analogue Addicts Phono was not designed for my own use. I was using an Artur Loesch derived Tube Phono at the time.
At the time members of an e-mail list (the Analogue Addicts) which was primarily England (not UK) based complained about the paucity of high quality and high performance Phono Stages in the sub 500 pound category and the fact that if there was a Phono Stage at all in your Amp or Pre, it was usually quite bad.
At the same time Audio DIY in the UK was going quite strong, in part at least helped by the original Maplin Shop chain.
Maplin carried a most amazing catalogue of electronic parts, including even OPA627 and 637 in metal cans and Philips Polystrene and Tinfoil capacitors, many excellent Audax Speaker Drivers (including the TW34 so beloved by the BBC) crossover components, the whole range of Velleman Kits including the much lusted after K4000 Tube Amp and their own range of Tube Kits (Amp and Preamp) supported by very high quality transformers and chokes also sold separate.
I remember, the set of Iron (Mains TX and Outputs) for a Stereo EL34 Amp (e.g. Dynaco ST-70 copy) could be bought for under 100 Pound.
Anyway, in this particular situation people that had easy access to parts, some modest experience messing with electronics and a great desire to build a high performance Phono.
So I decided to offer an older design of mine, adapted to use parts available from Maplin (and Cricklewood Electronics for the 2SK147) and keeping things very simple, to ensure a high success rate when building the design with limited tools and on a vero-board.
The choice of the choke and of batteries as power source was part of this. For people who did not want to use batteries the same chokes where originally recommended to be put into the main PSU Lines following a simple Vellemann 317/337 based PSU KIT.
If you where not a member of the Analogue Addicts E-Mail list in the mid 90's and did not live in England, the Analougue Addicts Phono was not designed for you.
It was most emphatically not designed for easy and low cost implementation by chinese Kopy-Mao's to turn a profit or to please some technocrat who will complain that it could have all been done at halve price of the choke with around 100 cheap parts using some kind of super hyper mega regulator, which in actual terms would still have had higher noise anyway ...
Ciao T
Most HP instruments from the '60s through the '80s have a lot of adjust on test points to make them meet spec. Sometimes its unavoidable with real parts.
A precision crystal oscillator only generates "noise" at its operating frequency (and harmonics). To a power supply its load fluctuates as much as a wirewound resistor (not much). However, power supply noise will show up as phase noise. And oscillators at the level of the Wenzels are so good to start with that even a very low noise regulator can limit the ultimate performance. The close in phase noise is very sensitive to low frequency noise from the power supply.
This is T's point about the "universal perfect regulator" nonsense. If you need 1 nV/rtHz at 50 Hz to meet the performance requirement then output Z may need to suffer.
I still think this is confusing, if you say ripple eater = gyrator.
Do you speak about John Curl??
He more than once said he uses a cap multiplier as the last bit of the power supply. Are you sure he uses "finesser" and "ripple eater", which are a different thing?
Demian,
If you have some Zout on the supply, wouldn't the oscillator's high speed switching cause appreciable noise on the supply line?
That isn't addressed at all by the Finesse.
jan
- Status
- Not open for further replies.