After resolving the problem with resistor oscillation (mod #2), we going further. Let’s try to receive the best power supply for R2R as possible.
Introduction
First of all I have modeled R2R with 4-levels only. The main specific of this R2R from PSU point of view is:
When is 0000 the ladder doesn't draw any current. And if there is at least one digit 1 a power draw is relatively much. A power consumption depends from an amount of '1's and its allocation. it leads to not equable power consumption. Each ladder switching creates a voltage slump and load response further. Please see below.
Below you can see the switching process influence on power supply (half wave measurement, 200Hz, 5mV scale)
The ripple inside the red circle is the ladder switching.
Anyway, to provide a clear power supply you have to have the following:
- a good regulator that has good noise parameters from one side and steadiness to load from another side.
- a good power supply capacity (many uFs)
mod #2 mk2
After several transistor based schemes, I came to a final one. And from my point of view it looks the best one in my oscilloscope. Even better than Soren’s mod proposition.
The last scheme
I used the original buffer scheme, but modify it to become as a pure follower.
Sequence of steps in the pictures
step 1. I simply removed the feedback resistor and set the shortage instead of the feedback capacitor. Removed parts are indicated:
step 2. Cutting the middle leg of the transistor.
step 3. Transistor soldering
step 4. the same approach for all arms.
step 5. Big resistor is added (2W)
step 6. The final test view is
Note the transistor collectors connected directly to the power supply units (+-5V).
Results of proposed scheme
Previous stage. Just without any additional capacitors in one channel.
This result achieved between positive and negative rails of registers (not between rail and gnd!!!), 2mV/div:
70Hz
200hz
12 Khz
The average noise level is 0.5-0.8 mV through all reachable frequencies.
Preliminary outcome:
This is the power supply for the resistor ladder which does not depend from the frequency of a signal. And its behavior almost identical throughout the audio range. Some signal smoothing marked in high frequencies is related to the small capacitor from original buffer scheme. In my opinion, any frequency-dependent parameters of the system are harmful to the sound. A frequency independent power supply means that we will get less intermodulation distortions and better tonal balance.
Comparison
So let’s add the capacitors and compare the different mods.
The stock Soekris DAC (output opAmps were removed as well)
600Hz
and
12 Khz
The last Soren’s mod proposition (220uf+220uf per channel and output opAmps were removed)
1500hz
and
12 Khz
And my last transistor approach (220uf+220uf per channel and output opAmps were removed as well)
12 Khz
If we look more precisely on the last screen, you can see the little ripple on the result line. After several attempts of adding the capacitors, i have found out that ~1000uf per channel giving the best results (see the schema in the beginning).
12 Khz
the similar view you can see everywhere through any frequencies in case of 1000uF(or more) using. The noise level is not more than 0.1mV.
Outcomes:
In the final, I managed to get the clearness of power supply, as I originally wanted. First of all it is frequency-independent. Secondly, it has vanishingly small noise value (at least, for my oscilloscope).
Usually such results can be obtained using a quality LDO-regulator. But regulators are built on the principle of control error and its compensation through a control element. This is feedback approach. But, any feedback systems has some reaction time. Btw, in this thread was mentioned a problem of step load response. Any system based on feedback has transient effects that are leads to additional RF noise. For an example, look at a chart for one of today's best regulators LT3042. Do you need such voltage spikes?
My power supply - it's just a transistor + capacitor. No feedback, tracking chains and that sort of things. As a consequence, no delays in reaction. Just clean power.
The sound of mod #2 mk2 + mod #3 is better comparing to mod #2 + mod #3. It is step up. Everything going further and better. The difference is not dramatic as stock vs mod#2+mod#3, but anyway the difference is tangible.
What comes next?
relating to mod #2 mk2
- testing different variants of bypassing by film caps
- testing different types of capacitors (sanyo oscon and nichicon KZ)
- adding even more capacitors
relating to mod #X
- researching in the area of shift registers bypassing
- improving in power supply units (+-5v)
- improving clock power supply (3.3v)
- and many others mods)
Introduction
First of all I have modeled R2R with 4-levels only. The main specific of this R2R from PSU point of view is:
When is 0000 the ladder doesn't draw any current. And if there is at least one digit 1 a power draw is relatively much. A power consumption depends from an amount of '1's and its allocation. it leads to not equable power consumption. Each ladder switching creates a voltage slump and load response further. Please see below.
An externally hosted image should be here but it was not working when we last tested it.
Below you can see the switching process influence on power supply (half wave measurement, 200Hz, 5mV scale)
An externally hosted image should be here but it was not working when we last tested it.
The ripple inside the red circle is the ladder switching.
Anyway, to provide a clear power supply you have to have the following:
- a good regulator that has good noise parameters from one side and steadiness to load from another side.
- a good power supply capacity (many uFs)
mod #2 mk2
After several transistor based schemes, I came to a final one. And from my point of view it looks the best one in my oscilloscope. Even better than Soren’s mod proposition.
The last scheme
An externally hosted image should be here but it was not working when we last tested it.
I used the original buffer scheme, but modify it to become as a pure follower.
Sequence of steps in the pictures
step 1. I simply removed the feedback resistor and set the shortage instead of the feedback capacitor. Removed parts are indicated:
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
step 2. Cutting the middle leg of the transistor.
An externally hosted image should be here but it was not working when we last tested it.
step 3. Transistor soldering
An externally hosted image should be here but it was not working when we last tested it.
step 4. the same approach for all arms.
An externally hosted image should be here but it was not working when we last tested it.
step 5. Big resistor is added (2W)
An externally hosted image should be here but it was not working when we last tested it.
step 6. The final test view is
An externally hosted image should be here but it was not working when we last tested it.
Note the transistor collectors connected directly to the power supply units (+-5V).
Results of proposed scheme
Previous stage. Just without any additional capacitors in one channel.
An externally hosted image should be here but it was not working when we last tested it.
This result achieved between positive and negative rails of registers (not between rail and gnd!!!), 2mV/div:
An externally hosted image should be here but it was not working when we last tested it.
70Hz
An externally hosted image should be here but it was not working when we last tested it.
200hz
An externally hosted image should be here but it was not working when we last tested it.
12 Khz
The average noise level is 0.5-0.8 mV through all reachable frequencies.
Preliminary outcome:
This is the power supply for the resistor ladder which does not depend from the frequency of a signal. And its behavior almost identical throughout the audio range. Some signal smoothing marked in high frequencies is related to the small capacitor from original buffer scheme. In my opinion, any frequency-dependent parameters of the system are harmful to the sound. A frequency independent power supply means that we will get less intermodulation distortions and better tonal balance.
Comparison
So let’s add the capacitors and compare the different mods.
The stock Soekris DAC (output opAmps were removed as well)
An externally hosted image should be here but it was not working when we last tested it.
600Hz
and
An externally hosted image should be here but it was not working when we last tested it.
12 Khz
The last Soren’s mod proposition (220uf+220uf per channel and output opAmps were removed)
An externally hosted image should be here but it was not working when we last tested it.
1500hz
and
An externally hosted image should be here but it was not working when we last tested it.
12 Khz
And my last transistor approach (220uf+220uf per channel and output opAmps were removed as well)
An externally hosted image should be here but it was not working when we last tested it.
12 Khz
If we look more precisely on the last screen, you can see the little ripple on the result line. After several attempts of adding the capacitors, i have found out that ~1000uf per channel giving the best results (see the schema in the beginning).
An externally hosted image should be here but it was not working when we last tested it.
12 Khz
the similar view you can see everywhere through any frequencies in case of 1000uF(or more) using. The noise level is not more than 0.1mV.
Outcomes:
In the final, I managed to get the clearness of power supply, as I originally wanted. First of all it is frequency-independent. Secondly, it has vanishingly small noise value (at least, for my oscilloscope).
Usually such results can be obtained using a quality LDO-regulator. But regulators are built on the principle of control error and its compensation through a control element. This is feedback approach. But, any feedback systems has some reaction time. Btw, in this thread was mentioned a problem of step load response. Any system based on feedback has transient effects that are leads to additional RF noise. For an example, look at a chart for one of today's best regulators LT3042. Do you need such voltage spikes?
An externally hosted image should be here but it was not working when we last tested it.
My power supply - it's just a transistor + capacitor. No feedback, tracking chains and that sort of things. As a consequence, no delays in reaction. Just clean power.
The sound of mod #2 mk2 + mod #3 is better comparing to mod #2 + mod #3. It is step up. Everything going further and better. The difference is not dramatic as stock vs mod#2+mod#3, but anyway the difference is tangible.
What comes next?
relating to mod #2 mk2
- testing different variants of bypassing by film caps
- testing different types of capacitors (sanyo oscon and nichicon KZ)
- adding even more capacitors
relating to mod #X
- researching in the area of shift registers bypassing
- improving in power supply units (+-5v)
- improving clock power supply (3.3v)
- and many others mods)
Before everybody start adding transistors, please note that there is a reason I use the opa365, it have a Vos spec of max 0.2mV. Adding transistors out of the loop, the Vref will not track anymore, resulting in larger 2nd harmonic distortion. Not that it's a big problem in many opinions....
You can get same better vref performance by lowering the 10R series resistors, but I haven't determined the optimal value yet.... The opa365 actually seems pretty stable with low value of the resistors, as long as you have large capacitive loads.
One more thing, you can't have that small load resistors, you exceed the current (and power) capacity of the +-5V regulators. Unless you use some other +-5V regulators.
And I repeat, doing large and/or complicated modifications can easily damage your dam1021.
You can get same better vref performance by lowering the 10R series resistors, but I haven't determined the optimal value yet.... The opa365 actually seems pretty stable with low value of the resistors, as long as you have large capacitive loads.
One more thing, you can't have that small load resistors, you exceed the current (and power) capacity of the +-5V regulators. Unless you use some other +-5V regulators.
And I repeat, doing large and/or complicated modifications can easily damage your dam1021.
Last edited:
The main idea of the mod #2 mk2 is power supplying of the ladder without any tracking at all.
Of course, a big current is big problems. So, I using the external PSU for +/-5V. And, as showed before, the transistors powered directly from PSUs. Now power consumption is about 140 mA per channel. And the base of each transistor consuming 2.8 mA or less from buffer opAmp. This value approximately the same as ladder consumption from the original scheme.
This is not the light mod and for sure you have to have the "straight-arms" and some exp.
Hope you will add the balanced connection / multi-channel support?
Last edited:
alecm, first of all. thank you very much for a deep insight provided by you. as much as i like tweaking the way you have done, I also appreciate a rigid way of identifying the weakest points in a cost/benefit analysis. as far as i understand the main problem here is a load transient...
but then you have TPS7A470x chip which behaves a way more nicely, at least as far as i understand it (see p. 11 of [1]) that you may place before the dac. even if you leave 7xL05s untouched with their noise levels at 100uVs, add any cap to make it less severe and you could hit the similar performance target, wouldn't you?
[1] http://www.ti.com/lit/ds/symlink/tps7a47.pdf
but then you have TPS7A470x chip which behaves a way more nicely, at least as far as i understand it (see p. 11 of [1]) that you may place before the dac. even if you leave 7xL05s untouched with their noise levels at 100uVs, add any cap to make it less severe and you could hit the similar performance target, wouldn't you?
[1] http://www.ti.com/lit/ds/symlink/tps7a47.pdf
alcem, why not go all the way and get rid of the opamp - power the transistor from the reference? You simply need to determine how much base current the transistor needs, and make sure the reference can supply it - shouldn't be a problem.
Plus Soren is correct (of course) - this feedback free scheme will produce greater distortion due to lack of tracking. Worth keeping in mind when evaluating the sound. Also, a well implemented feedback reg will have superior load regulation than a feedback free follower. If you test a follower with a stepped load as the LT3042 that you show, it won't be pretty.
Plus Soren is correct (of course) - this feedback free scheme will produce greater distortion due to lack of tracking. Worth keeping in mind when evaluating the sound. Also, a well implemented feedback reg will have superior load regulation than a feedback free follower. If you test a follower with a stepped load as the LT3042 that you show, it won't be pretty.
Last edited:
Mod#2 and Mod#3 for raw output
Dear Alecm,
If I only use raw output, is there sound improvement after make mod#2 and mod#3?
Thanks
Comickitkit
Dear Alecm,
If I only use raw output, is there sound improvement after make mod#2 and mod#3?
Thanks
Comickitkit
In the mod description, I have described it.
Briefly again. Current consumption of base of the transistor is now about 2.8 mA. This compares with a buffer current. At the same time, Soren was take decision to feed the registers through the buffer and not directly from the inverter and Vref source. I cannot afford to ignore it. I believe that buffer removing from the circuit is potentially dangerous. At least - for power supply quality.
People, I'm still planning on doing two experiences when time permits:
1 - I'll try the bypass with mixed values and types of capacitors, influenced by the experiences of my friend who I quoted in some post ago. I confess that curiosity is killing me ... although I still reiterate that the effects here are low frequency as reported by alecm so should not to be so beneficial with the mix of various capacitance values.
2 - I'm thinking to build an double 4V shunt reg high precision and low impedance, in the same style of one I did for a pre phono, powered by a CCS, which would eliminate the need for internal of 5V and 4V supplies, assuming the risk of destroying my DAM ... a parallel possibilitade is to add some output transistor within the original regulator feedback loop like alecm, but with idea to giving power dissipation capatibilities to the regulator and allowing it to be CCS powered so automatically converting it in a shunt reg supply. The shunt will block the supply variations across the board. Even so I doubt if this will make so much difference, but since we are DIY...
Ah, very cool the new glt and More DAM filters blog posts!
1 - I'll try the bypass with mixed values and types of capacitors, influenced by the experiences of my friend who I quoted in some post ago. I confess that curiosity is killing me ... although I still reiterate that the effects here are low frequency as reported by alecm so should not to be so beneficial with the mix of various capacitance values.
2 - I'm thinking to build an double 4V shunt reg high precision and low impedance, in the same style of one I did for a pre phono, powered by a CCS, which would eliminate the need for internal of 5V and 4V supplies, assuming the risk of destroying my DAM ... a parallel possibilitade is to add some output transistor within the original regulator feedback loop like alecm, but with idea to giving power dissipation capatibilities to the regulator and allowing it to be CCS powered so automatically converting it in a shunt reg supply. The shunt will block the supply variations across the board. Even so I doubt if this will make so much difference, but since we are DIY...
Ah, very cool the new glt and More DAM filters blog posts!
Last edited:
You certainly wouldn't want to power the registers from the references, they wouldn't supply the required current. But using a buffer transistor as you have done would be OK. Only propbelm is the RC filter after the reference, you'll get some voltage drop there. Using a FET would solve that.
Personally, I don't think adding a transistor is necessary, tweaking the existing feedback reg will give very good results. Having said that, it seems you are already getting good results, so well done! Look forward to hearing more.
OOOps, another text problem
: The correct: a parallel possibilitade is to add some output transistor within the original regulator feedback loop UNlike alecm....The alecm is open loop and will be othe possibilitie for shunt testing, too.NOTE: the open loop emitter follower has a output impedance proportional to emitter current, of course. So with shunt supply we not need to use additional resistor loading the output to lowering their impedance, we only need to set the CCS feed with high current and voilà! Same effect and helps to isolating the shift register sector.
1. For sure i will investigate the 2nd harmonic distortion problem in transistor approach.
2. I have real and practical results that better than Soren's proposition. Please compare the screens in the same conditions.
3. Soren's mod has some limitation in capacitor expansibility as distinguished from transistor approach. You have a possibility to add 560-680uf maximum. Further, there is a chance to catch the oscillation according to http://www.ti.com/lit/an/sbva002/sbva002.pdf.
4. Regarding FET, Please draw a diagram on paper.
The main difference btw a pure emitter follower and shunt regulator is a load step response. Like any other regulator too. The main question is speed. The emitter response is smoothest.
Actuating even at MHz range, the OL emitter follower will be good for commutating circuits.
You can use the emitter follower in an open loop shunt regulator, too. I used one simple shunt version in my LatMOS ZEN I/V, please check : http://www.diyaudio.com/forums/pass-labs/173291-zen-i-v-converter-37.html (and forgive my BAAAD hand draw schematic...)
With good results. For DAM we can use the original internal reference like in your series supply version, is far better reference than simple zener, of course... but the R2R voltage then become +/-4.65V instead of 3.35V from "series" emitter follower.
Well, this ZEN I/V project I made before the DAM acqusition (but posted recently to people know about, since varios people have DACs to use them). I can say the PMD100 digital filter sounds good and I have to using the most recent filter in DAM to superate the 44100Hz performance from this tweaked EAD DAC.
Do you mean this ?
An externally hosted image should be here but it was not working when we last tested it.
Yes, and with R3 and R4 being CCS's, but need more than <1V disponible to make an good CCS, so they will need to be connected to 12V.
We can experiment with resistors if the predecessor supply is clean and appreciate the results.
Last edited:
I see the main advantage of this approach is load step eliminating for +/-5 psu. it dramatically reducing the implementation requirements.The main disadvantage is heat dissipation on transistors. But big caps and big current will be anyway.
So, this is an interesting idea. Maybe I will try it later.
An other approach to keep Vref stable, would be to keep the load stable. This could e.g. be done by using one DAM per cannel, feeding the second ladder with the complementary signal (in sense of current load) and use a common Vref for both ladders. This is an a bit "wasteful" approach, but should, in theory, give an a much cleaner Vref.