Sorry for my english in advance , it is not my native language. As I have promised before, want to share my experience in Soekris mods. The main idea was to retry the bravest mods from the thread. But initially want to describe my assessment of sound quality for non-modified Soekris DAC. The main advantages of Soekris DAC (not modified) are (compared with the Esoteric D-07): timbres; fantastic mids; good bass. The disadvantages are: sound and instruments are sticking to loudspeakers. Instead of dissolving in a room the sound comes directly from the speakers (in my experience this is an obvious problem of power supply); scene is flat without depth; moderate dynamics; localization problems on a stage.
Mod #1
First little mod related to 3.3V source and it was inspired by datasheet of ZLDO1117. The idea to exchange 4.7uF by 10uF tantalum cap.
The result from tantalum cap 10uF looks a little bit smoother. Below you can see how it looks like.
Unfortunately, I couldn’t see the difference in the cap changing; it is beyond borders of my oscilloscope. But, at least it not become worse - the curves on the oscilloscope before and after replacing looks more or less identically.
Mod #2
After listening session, I took decision to improve the power supply of the most critical DAC part (from the first point of view). This is the power supply of R-2R ladder. Obviously, the resistors can’t be fed independently. They are powered through the outputs of the shift registers LVC595. So I analyzed the power supply of these registers in a first raw.
Here you can see the Vcc leg оf LVC595 chip.
I set the pure sine 20 kHz signal to the DAM input and caught it in the DAM buffered output over the second channel of oscilloscope.
The scale for this curve is 2 V/div. You can see the pure input sine on the bottom of screen. In the top, you can see the output sine captured from Vcc point. And it looks very bad taking into account that it going to the ladder directly and for sure it has a negative influence over the music quality. AC is strongly correlated to the main signal and if the input signal will be the 12 kHz the output signal has to be the same on the time axis.
By the way, this strange distortion of power supply has the maximum value on 12 kHz. The signal has less distortion level on other frequencies. And it doesn’t depend from a clock frequency: 48, 96, 192 and 384 has the same behavior. The main question is why?
The next analyzing point was the source of +-5V.
The curve is very similar to previous one but scaled up even more. 7xL05 are not the best regulator, possibly this is the reason of increasing.
Let’s start the mod and desolder original +-5V regulators. See the result below.
Further, let’s connecting the source of clear 5V, based on the fresh LT regulator (LT3042 0.8uV Ultralow noise DAC power supply regulator 3.3V/5V 1.5A*x2 - DIYINHK). It has the best noise parameter through current regulators, less than 1uV! The wires connected directly on the same traces of former 7xL05 regulators.
let’s do test it again. And….. the first bad surprise. +4V is not changed significantly and almost the same with original one. wth?
More deep analysis has revealed additional details. Some miracle circuit is generating the reference voltage of -4V. Further, this voltage inverting to the +4V by an opamp:
Next, there is one more opAmp used as a buffer and this is the BB OPA365.
What is the most interesting that direct buffer output has a CLEAR +4V (see ‘clear’ hints above)!! The mentioned distortions are raising somewhere later.
For the reason of stability a capacitive loading (this is the common problem of opAmps) usually using some RC circuit. Here we have R = 10 Ohm (see the red circles below) and C = don’t know (but it doesn’t matter in this particular case).
The source of mentioned distortion is 10 Ohm resistor!!! It works not for the main purposes only, but as **** generator additionally.
To avoiding the problems of mentioned resistor, I propose to exchange it by transistor. The main idea of the mod is to provide a stronger power source. The simplest way is to use an emitter follower. These are the simplified schematics, original one and proposed slightly below:
The side effect of transistor using is a voltage drop. In our particular case, I used NXP BC850 (NPN) and BC860 (PNP) transistors and having 0.6V drops on the base-emitters. As a result, we have 3.4V only, instead of 4V. I hope this is not the problem for the end result and sound quality.
4 resistors were removed and on the same places were installed 4 transistors. Collectors connected directly to +/-5V decoupling capacitors (see 4 white wires). Additionally, each installed transistor was bypassed by 220uF polymer capacitor, but the final capacitor size is the subject of future experiments.
So let’s see the final result
and original one.
The mod#2 result is good and we have definitely more clear Vcc and power supply of R2R ladder accordingly. On the 12 kHz, we have 6 mV distortion level in the original one and 1mV only for mod#2.
Mod #3
this mod the same as #7 from http://www.diyaudio.com/forums/vend...magnitude-24-bit-384-khz-292.html#post4369503
The reason is the 7 nA current consumption of LME49710 from RAW output. It is too high in comparison to overall consumption level. And this level is comparable to the signal current on 8 least significant bits.
Outcome
The mod#1 result is not auditioned in details. But the results in sound of last two mods are amazing in some areas. The most significant improvements are:
- Very deep soundstage (mod#2) outward room bounds with good localization (mod#3).
- The loudspeakers are dissolved in the room (both mod#2 and #3)
- Tremendously increased sound dynamics (both mod#2 and #3)
- Significant increased resolution and detailization (both mod#2 and #3, but especially mod#2)
The result is not final and will be examined further. Unfortunately some old rock records has some sound distortion in the midbass area.
Mod #1
First little mod related to 3.3V source and it was inspired by datasheet of ZLDO1117. The idea to exchange 4.7uF by 10uF tantalum cap.
An externally hosted image should be here but it was not working when we last tested it.
The result from tantalum cap 10uF looks a little bit smoother. Below you can see how it looks like.
An externally hosted image should be here but it was not working when we last tested it.
Unfortunately, I couldn’t see the difference in the cap changing; it is beyond borders of my oscilloscope. But, at least it not become worse - the curves on the oscilloscope before and after replacing looks more or less identically.
Mod #2
After listening session, I took decision to improve the power supply of the most critical DAC part (from the first point of view). This is the power supply of R-2R ladder. Obviously, the resistors can’t be fed independently. They are powered through the outputs of the shift registers LVC595. So I analyzed the power supply of these registers in a first raw.
Here you can see the Vcc leg оf LVC595 chip.
An externally hosted image should be here but it was not working when we last tested it.
I set the pure sine 20 kHz signal to the DAM input and caught it in the DAM buffered output over the second channel of oscilloscope.
An externally hosted image should be here but it was not working when we last tested it.
The scale for this curve is 2 V/div. You can see the pure input sine on the bottom of screen. In the top, you can see the output sine captured from Vcc point. And it looks very bad taking into account that it going to the ladder directly and for sure it has a negative influence over the music quality. AC is strongly correlated to the main signal and if the input signal will be the 12 kHz the output signal has to be the same on the time axis.
By the way, this strange distortion of power supply has the maximum value on 12 kHz. The signal has less distortion level on other frequencies. And it doesn’t depend from a clock frequency: 48, 96, 192 and 384 has the same behavior. The main question is why?
The next analyzing point was the source of +-5V.
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.
The curve is very similar to previous one but scaled up even more. 7xL05 are not the best regulator, possibly this is the reason of increasing.
Let’s start the mod and desolder original +-5V regulators. See the result below.
An externally hosted image should be here but it was not working when we last tested it.
Further, let’s connecting the source of clear 5V, based on the fresh LT regulator (LT3042 0.8uV Ultralow noise DAC power supply regulator 3.3V/5V 1.5A*x2 - DIYINHK). It has the best noise parameter through current regulators, less than 1uV! The wires connected directly on the same traces of former 7xL05 regulators.
An externally hosted image should be here but it was not working when we last tested it.
let’s do test it again. And….. the first bad surprise. +4V is not changed significantly and almost the same with original one. wth?
An externally hosted image should be here but it was not working when we last tested it.
More deep analysis has revealed additional details. Some miracle circuit is generating the reference voltage of -4V. Further, this voltage inverting to the +4V by an opamp:
An externally hosted image should be here but it was not working when we last tested it.
Next, there is one more opAmp used as a buffer and this is the BB OPA365.
An externally hosted image should be here but it was not working when we last tested it.
What is the most interesting that direct buffer output has a CLEAR +4V (see ‘clear’ hints above)!! The mentioned distortions are raising somewhere later.
For the reason of stability a capacitive loading (this is the common problem of opAmps) usually using some RC circuit. Here we have R = 10 Ohm (see the red circles below) and C = don’t know (but it doesn’t matter in this particular case).
An externally hosted image should be here but it was not working when we last tested it.
The source of mentioned distortion is 10 Ohm resistor!!! It works not for the main purposes only, but as **** generator additionally.
To avoiding the problems of mentioned resistor, I propose to exchange it by transistor. The main idea of the mod is to provide a stronger power source. The simplest way is to use an emitter follower. These are the simplified schematics, original one and proposed slightly below:
An externally hosted image should be here but it was not working when we last tested it.
The side effect of transistor using is a voltage drop. In our particular case, I used NXP BC850 (NPN) and BC860 (PNP) transistors and having 0.6V drops on the base-emitters. As a result, we have 3.4V only, instead of 4V. I hope this is not the problem for the end result and sound quality.
4 resistors were removed and on the same places were installed 4 transistors. Collectors connected directly to +/-5V decoupling capacitors (see 4 white wires). Additionally, each installed transistor was bypassed by 220uF polymer capacitor, but the final capacitor size is the subject of future experiments.
An externally hosted image should be here but it was not working when we last tested it.
So let’s see the final result
An externally hosted image should be here but it was not working when we last tested it.
and original one.
An externally hosted image should be here but it was not working when we last tested it.
The mod#2 result is good and we have definitely more clear Vcc and power supply of R2R ladder accordingly. On the 12 kHz, we have 6 mV distortion level in the original one and 1mV only for mod#2.
An externally hosted image should be here but it was not working when we last tested it.
Mod #3
this mod the same as #7 from http://www.diyaudio.com/forums/vend...magnitude-24-bit-384-khz-292.html#post4369503
The reason is the 7 nA current consumption of LME49710 from RAW output. It is too high in comparison to overall consumption level. And this level is comparable to the signal current on 8 least significant bits.
Outcome
The mod#1 result is not auditioned in details. But the results in sound of last two mods are amazing in some areas. The most significant improvements are:
- Very deep soundstage (mod#2) outward room bounds with good localization (mod#3).
- The loudspeakers are dissolved in the room (both mod#2 and #3)
- Tremendously increased sound dynamics (both mod#2 and #3)
- Significant increased resolution and detailization (both mod#2 and #3, but especially mod#2)
The result is not final and will be examined further. Unfortunately some old rock records has some sound distortion in the midbass area.
Very interesting work Alecm. The inverter OPAMP is not OPA365 though. The OPA365s drive the shift registers as you have correctly pointed out. The inverter circuit uses a TS507 I think? The -4V reference is based on a TL431.
What is the real scale of your upper oscilloscope traces? I don't think it really can be 2v/div?
What is the real scale of your upper oscilloscope traces? I don't think it really can be 2v/div?
Last edited:
Nice work
I would say that the 220u polymer capacitors have way more effect than the transistors, so just putting a large polymer capacitor in parallel with each of the +-4V ceramics should have same effect.
Mod #1 don't do anything, there are already plenty of large and small ceramic capacitors on the 3.3V supply, totally around 65 uF.
wineds, thanks, hope you will find my work useful to retry.
May be we have different board versions? Both opamps has the same marking - OAVQ
thanks for TL431.
According to scale, the top wave has 2mV per div, and the bottom wave has 2V per div
Nice work
I would say that the 220u polymer capacitors have way more effect than the transistors, so just putting a large polymer capacitor in parallel with each of the +-4V ceramics should have same effect.
Soren, thanks
A transistor base is easy load for any opamps and you have free hands to increase capacitor values without any fears to become unstable, leading to oscillation.
For sure, there is very interesting to try caps in parallel (bypass). possibly someone try it as well and will share his exp.
My idea was not using large values but ESR cap parameters more.
Dear Alecm,
Thanks for your sharing on the DAM. But I am not clear that for LME49710 removal from RAW output. is it the works just only to remove it from the board and don't need to do anything further on the mod#3 at your post or need to do the connection some points to complete the circuit?
thanks so much
Thanks for your sharing on the DAM. But I am not clear that for LME49710 removal from RAW output. is it the works just only to remove it from the board and don't need to do anything further on the mod#3 at your post or need to do the connection some points to complete the circuit?
thanks so much
Interesting. Mine is marked K136. Clearly there are two versions of this board. Maybe only the boards loaded with a TS507 have the +4V startup issue.
alecm,
This schematic is incorrect :
http://s5.postimg.org/qyxagrb7b/014.png
The 10 ohm is inside the OPAMP feedback loop. The feedback resistor (marked as 68A) is 500 ohm.
This schematic is incorrect :
http://s5.postimg.org/qyxagrb7b/014.png
The 10 ohm is inside the OPAMP feedback loop. The feedback resistor (marked as 68A) is 500 ohm.
This is a very good observation. Congratulations with the work and the results. It is obvious 6 mV ripple is not meant to be and that it has influence on the outcome. I hope Soekris will change the design to your findings.
If you are referring to the supplying the +/-4V externally, it needs to be understood that both rails need to be exactly the same magnitude. Otherwise distortion will be introduced. A tracking power supply is required.
hi, comickitkit,
this is the very interesting point.
IMHO, to receive the maximum result you have to remove it anyway. But implementing mod#3 only will not help you enough. The good result I have received to combine both of it. And this is the key. When i have done the mod#2 and still not have done #3, i received the bad result in stage and localization point of view. The sound looked like out of phase. The mod#3 fully resolved it.
The raw output looks like a "raw nerve". It critically vulnerable from everything what are sticking around and afterwards especially! So if you done some mods but not received any good results or received a strange result, check the next components after Soekris dac as well! I have top preamplifier from MARANTZ (Marantz US | SC-7S2) and everything looks good. But when I had been testing the mods during the soldering process, I have been receiving the very bad results(sound full of distortions and rattling) from RAW output, because i used a little active PC speakers connected to RAW.
One more remark. I removed both the 49710s and the 49724s. But, removing 49710 only and leaving 49724 on the board is mistake. The 49724 inputs will fly and opAmp will produce noise to power rails. So, if you took the decision to remove, remove their both.
Yes,
TS507 is pin-to-pin compatible with OPA365. Not good for buffer, but good enough for voltage inverter (at least, 1.9MHz vs 50MHz). So, OPA365 and TS507 can substitutes each other in this place.
wineds,
For sure, this is not the exact schema but simplified one, just for illustrating a transistor idea. In the reality, there is resistor there. For the simplification reasons, I connected the output with negative input.
56uf. That's all I had at hand. I need to order some 1X scope probes to measure the outcome...
hi glt,
first of all, big thanks for the great resource - https://hifiduino.wordpress.com/!
You hit directly to the one of my next mod ideas! For sure it is reasonable to add caps to each shift register. I'm still in the thinking process, but tending to use polymers.
Instead of using a one big cap near buffer considering the option to set caps with lesser values additionally for each ceramic cap. Total register amount is 16 or 8 per rail. Thus, we get the same total capacity with smaller dimensions. But in any case, to save the overall capacity has to be taken at least 22uF. So I would advise using a quality electrolytes or polymers from 33 to 100 uF.