Hi Okapi, yes, sure. I'm just eagerly rediscovering my favorite recordings, many of which are original masters, and am absolutely stunned by what I'm hearing. Goosebumps! If your speakers have good phase behavior you will be blown away by the naturalness and holography.
LM317:
out to adjust - 240 ohms
adjust to gnd : 2.7k ohms || 0.12uF
output cap: 1500uF w/ 10uF film
minimum load: 100mA
LM337:
out to adjust - 240 ohms
adjust to gnd : 2.7k ohms || 0.033uF
output cap: 1000uF w/ 10uF film
minimum load: 50mA
Input to output differential : 2.5 volts
LM317:
out to adjust - 240 ohms
adjust to gnd : 2.7k ohms || 0.12uF
output cap: 1500uF w/ 10uF film
minimum load: 100mA
LM337:
out to adjust - 240 ohms
adjust to gnd : 2.7k ohms || 0.033uF
output cap: 1000uF w/ 10uF film
minimum load: 50mA
Input to output differential : 2.5 volts
Sorry, I still use the old English system for parts values notation. I've edited it for clarity.
Noise is worse, no doubt. I'll measure it tomorrow. But after hearing this, I have to say that noise is A LOT less important than I have thought previously. Give me impedance balance any day.
Surely I am not the first person to think of this.
Notice also the impedance phase is pretty linear up to 5 kHz too.
There are some really bright engineers on this site that are alot smarter than me in electronics. I'd love to see one of them develop a balanced-impedance bipolar supply that was down below say .005 ohms or less, and with even a tighter Z balance over the spectrum. Maybe Jonathan Carr will bite. I knew him in the past and his attention to detail is wonderful.
Alot of the evils that have been ascribed to solid state can be traced back to this, methinks. I can't wait to do my phono stage power supply. This should be stunning for high-gain circuits!
Noise is worse, no doubt. I'll measure it tomorrow. But after hearing this, I have to say that noise is A LOT less important than I have thought previously. Give me impedance balance any day.
Surely I am not the first person to think of this.
Notice also the impedance phase is pretty linear up to 5 kHz too.
There are some really bright engineers on this site that are alot smarter than me in electronics. I'd love to see one of them develop a balanced-impedance bipolar supply that was down below say .005 ohms or less, and with even a tighter Z balance over the spectrum. Maybe Jonathan Carr will bite. I knew him in the past and his attention to detail is wonderful.
Alot of the evils that have been ascribed to solid state can be traced back to this, methinks. I can't wait to do my phono stage power supply. This should be stunning for high-gain circuits!
i think an important factor to keep in mind when trying to duplicate jbau's results is that they should be dependent on capacitor ESR.
An alternative method to measuring Zout and perhaps more accessible is to do a transient load test and measure the ringing frequency. The adjust cap value can be calculated based on the ringing frequency.
jbau, i think this has been an excellent exercise in developing a power supply with a flat impedance profile across the audio band. Since you are pretty sure that the effect of the flat Zout of the power supply is audible in your system, is there a way you can back up your subjective observation with some measurement? Can you measure the THD, IMD, and or noise floor of the device you put the power supply in? Can you measure these same variables with the original power supply so you have a proper control? I think it is worth the effort to try and distinguish your result from the hundreds of other unverified claims found in audioland.
An alternative method to measuring Zout and perhaps more accessible is to do a transient load test and measure the ringing frequency. The adjust cap value can be calculated based on the ringing frequency.
jbau, i think this has been an excellent exercise in developing a power supply with a flat impedance profile across the audio band. Since you are pretty sure that the effect of the flat Zout of the power supply is audible in your system, is there a way you can back up your subjective observation with some measurement? Can you measure the THD, IMD, and or noise floor of the device you put the power supply in? Can you measure these same variables with the original power supply so you have a proper control? I think it is worth the effort to try and distinguish your result from the hundreds of other unverified claims found in audioland.
AndrewT said:
although a tracking pre reg would have the advantage of keeping the in/out voltage differential constant across the output regulator i think it could also change the output impedance. I think that the Zout of the second regulator in the tracking regulator would in part be determined by the first regulator whenever the first regs Zout exceeds the second regulators Zout. I think it could be done but it would make tuning of the circuit a bit more difficult. Is it worth it? I'm not so sure, the frequency region over which the in/out differential effects Zout is pretty narrow and it's effect is not that large.
okapi: I don't think the ESR is as deterministic as you suggest. Just use reasonable good quality parts and you should be ok.
I'm highly skeptical that transient load test will get you to the same place. It focuses on one phenomena in the presence of several interactive ones. And it will be useless when matching the 337 characteristics to the 317, since the former exhibits no resonances. But it's clear we need to come up with a testing method that can be done with less esoteric equipment. Perhaps a current probe would be useful.
To remove my work from the dac and reinstall the old components sounds like a school assignment. I have another nearly identical dac I'll be modding soon so I'll do some measurements to compare then. This is a simple enough circuit to build; put one together and listen.
To say "you are pretty sure that the effect of the flat Zout of the power supply is audible in your system" is the understatement of the century. It's been several years since I was last moved to tears by listening to recordings.
I would love to hear the contrary reasoning; why can flat, low, and balanced Z be ignored and how can it NOT have a significant audible impact? I think my view is complete logical. Supply impedance variations will affect levels of current delivered. Phase variations will affect timing relationships. How can it be otherwise?
I think that the power system impedance has been ignored because it is difficult to measure and would be a major nuisance to manufacture consistently. Just like phase response in speakers was ignored for decades for the same reasons.
I'm highly skeptical that transient load test will get you to the same place. It focuses on one phenomena in the presence of several interactive ones. And it will be useless when matching the 337 characteristics to the 317, since the former exhibits no resonances. But it's clear we need to come up with a testing method that can be done with less esoteric equipment. Perhaps a current probe would be useful.
To remove my work from the dac and reinstall the old components sounds like a school assignment. I have another nearly identical dac I'll be modding soon so I'll do some measurements to compare then. This is a simple enough circuit to build; put one together and listen.
To say "you are pretty sure that the effect of the flat Zout of the power supply is audible in your system" is the understatement of the century. It's been several years since I was last moved to tears by listening to recordings.
I would love to hear the contrary reasoning; why can flat, low, and balanced Z be ignored and how can it NOT have a significant audible impact? I think my view is complete logical. Supply impedance variations will affect levels of current delivered. Phase variations will affect timing relationships. How can it be otherwise?
I think that the power system impedance has been ignored because it is difficult to measure and would be a major nuisance to manufacture consistently. Just like phase response in speakers was ignored for decades for the same reasons.
The supply noise measurements are surprising. For the 317 at 100mA load current, it is just shy of 300uV in a 10Hz-30kHz bandwidth. Swapping in a 10uF adjust cap (as per factory) is 290uV. Not a big difference.
With a 400Hz hipass engaged, noise is actually 2dB higher with the 10uF cap compared to the 33nF. This infers that ripple rejection under load conditions is actually worse with the 10uF cap.
So my hunch was wrong. There is no noise price to be paid when optimizing the 317 for flattest Z. Light loading gives other results. But once it's in it's linear current zone, a larger adjust cap has little to negative impact on noise.
With a 400Hz hipass engaged, noise is actually 2dB higher with the 10uF cap compared to the 33nF. This infers that ripple rejection under load conditions is actually worse with the 10uF cap.
So my hunch was wrong. There is no noise price to be paid when optimizing the 317 for flattest Z. Light loading gives other results. But once it's in it's linear current zone, a larger adjust cap has little to negative impact on noise.
jbau said:
Doesn't this imply just the opposite?
My thinking: With 10hz to 30khz noise was about equal (290 vs 300). When you remove ripple frequencies (look above 400hz) from the measurement noise is now higher with the 10uF cap. Doesn't this mean that with the 10 uF cap there was less ripple/noise below 400hz than with the 33nF cap?
Huh? With the 400Hz filter in, the noise with the 10uF cap is 2dB higher than it is with the 33nF cap. What's not clear here? Some of the noise gain is above 400Hz. We saw that with the 2.5kHz peak before, remember? The 10uF cap notches out stuff centered on approx. 600 Hz. Maybe my wording wasn't very clear.
gimme a chance to understand.
with the 400 hz filter in you are measuring the noise above 400hz (400 hz to 30khz i assume).
you see more noise with the 10uF cap in this case.
however, when you looked from 10hz to 30 khz you saw essentially equal noise.
doesn't this mean from 10hz to 400hz the 10uF cap outperformed the 33nF
my reasoning assumes that the noise measurement is taking an average of the noise across the bandwidth of the measurement and not the max noise.
with the 400 hz filter in you are measuring the noise above 400hz (400 hz to 30khz i assume).
you see more noise with the 10uF cap in this case.
however, when you looked from 10hz to 30 khz you saw essentially equal noise.
doesn't this mean from 10hz to 400hz the 10uF cap outperformed the 33nF
my reasoning assumes that the noise measurement is taking an average of the noise across the bandwidth of the measurement and not the max noise.
Yes technically you're right but I was remembering the noise sweeps early in this series where the cap didn't affect stuff below 150-200Hz but bumped the noise at 2kHz. The cap affects approx 200Hz to 3kHz or so. The Vin/Vout diff affects say 800Hz and below. The resistor parallel impedance value affects the entire range up to 10kHz or so. The output cap shapes things above about 2kHz. That's the ingredients in this stew. 
I need to look back at those noise measurements but weren't they taken before you discovered you had a really big in/out voltage differential?
now i need to try and give your previous post a worthy response.
Here i was referencing the article by Erroll Dietz. In his article he is talking about the noise peaks and shows how it scales with ESR. I assumed that this noise peak was the same thing that caused the impedance dip. i may be incorrect here.
here i was referencing the article you posted earlier:
http://www.ema-eda.com/products/other/articles/Regulator2.pdf
i think you are definitely right in circumstances where you don't have an oscillation.
is there an economical current probe available. the tektronix ones i see on ebay are quite expensive?
the scientist in me forces me to say: changing more than one variable at a time invalidates this experiment. However i certainly understand your motivation.
i think the hypothesis your present is logical but you have not presented any empirical evidence using established techniques to verify your observation.
i have a alternative hypothesis that may be wrong because it depends on what the PS is powering but i am going to assume it is powering all parts of the DAC. Here goes: A DAC has digital components that operate at frequencies much higher than those you optimized the DAC for. To obtain the best response over the chosen bandwidth you likely made sacrifices at higher frequencies, frequencies that matter in the digital domain.
i could not agree more. i don't think it is realistic for a large volume manufacturer to do this type of work.
now i need to try and give your previous post a worthy response.
jbau said:
Here i was referencing the article by Erroll Dietz. In his article he is talking about the noise peaks and shows how it scales with ESR. I assumed that this noise peak was the same thing that caused the impedance dip. i may be incorrect here.
here i was referencing the article you posted earlier:
http://www.ema-eda.com/products/other/articles/Regulator2.pdf
i think you are definitely right in circumstances where you don't have an oscillation.
is there an economical current probe available. the tektronix ones i see on ebay are quite expensive?
the scientist in me forces me to say: changing more than one variable at a time invalidates this experiment. However i certainly understand your motivation.
i think the hypothesis your present is logical but you have not presented any empirical evidence using established techniques to verify your observation.
i have a alternative hypothesis that may be wrong because it depends on what the PS is powering but i am going to assume it is powering all parts of the DAC. Here goes: A DAC has digital components that operate at frequencies much higher than those you optimized the DAC for. To obtain the best response over the chosen bandwidth you likely made sacrifices at higher frequencies, frequencies that matter in the digital domain.
i could not agree more. i don't think it is realistic for a large volume manufacturer to do this type of work.
okapi, a response to the things we didn't address privately:
Tek current probes: The expensive ones are the ones that are in demand in the industrial market, like the AM503S systems, the higher-bandwidth (60MHz) P602x probes, etc. The previous generation were 20MHz and are just fine for audio use with their passsive terminations into normal voltage preamps. You can usually pick those up for cheap. I use an old P6016 with passive term and it works great. Just get to know it's level and freq response so you know it's characteristics and subtract that out of the measurements. A network analyser does that for you automatically. Or you can apply a correction curve in post processing if you are digitizing. That way you can still get useful information from them outside of their normal passband.
My DAC is a DAT recorder, it has separate regulators for everything, so the 15V rails only drive audio IC's and other regulators and it's ok to tailor it for audio. The grounding scheme on it is pretty good (Sony was ahead of the curve on that count).
Tek current probes: The expensive ones are the ones that are in demand in the industrial market, like the AM503S systems, the higher-bandwidth (60MHz) P602x probes, etc. The previous generation were 20MHz and are just fine for audio use with their passsive terminations into normal voltage preamps. You can usually pick those up for cheap. I use an old P6016 with passive term and it works great. Just get to know it's level and freq response so you know it's characteristics and subtract that out of the measurements. A network analyser does that for you automatically. Or you can apply a correction curve in post processing if you are digitizing. That way you can still get useful information from them outside of their normal passband.
My DAC is a DAT recorder, it has separate regulators for everything, so the 15V rails only drive audio IC's and other regulators and it's ok to tailor it for audio. The grounding scheme on it is pretty good (Sony was ahead of the curve on that count).
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.