Andrew,
I agree that the colors are hard to make out. As a pdf, I zoomed way in on a large screen and was able to correlate the colors to the legend in almost all cases, but it was difficult.
The author doesn't tell us much other than the model. I looked up many of the datasheets, especially on the low noise side. The vast majority of these are metal thin film types. There are one or two metal oxides and only one carbon film that I can remember. The interesting thing about the carbon film is that it is below the -40 dB line at 100 Ohms. However, it is among the worst of that grouping.
I agree that all of the resistors approach Johnson noise level at higher frequencies. For 100 ohm resistors up to 1 W, both SMD (Fig 6) and through hole (Fig 7), the lowest noise resistors approach Johnson noise level in the neighborhood of 100 Hz. The poorer group are all thick film with one bad thin film in the bunch. They get to Johnson noise level at higher frequencies.
At 1 kOhms and 10 kOhms, the best resistors reach Johnson noise level at similar frequency to their 100 Ohm cousins. Also, the higher power resistors reach Johnson noise at higher frequencies. The highest power group tested was all thick film design which is expected to have higher noise.
Note that in Figure 7, the author tested a bulk film resistor, Vishay S102K, which of course had the lowest noise.
I guess my point is that resistor manufacturers are continuing to improve. There are many resistors with lower than -40 dB noise and some quite a bit better. With the exception of the bulk film resistor and some parts from small makers that I can't find information on, everything here is from an industrial or military spec resistor. Many cost the same as resistors we are using now. Also, this study was from 2009. In the intervening 8 years, some of these resistors have become obsolete and replaced with better performing models.
If you can build your project with lower noise components for the same money, why not do it?
I agree that the colors are hard to make out. As a pdf, I zoomed way in on a large screen and was able to correlate the colors to the legend in almost all cases, but it was difficult.
The author doesn't tell us much other than the model. I looked up many of the datasheets, especially on the low noise side. The vast majority of these are metal thin film types. There are one or two metal oxides and only one carbon film that I can remember. The interesting thing about the carbon film is that it is below the -40 dB line at 100 Ohms. However, it is among the worst of that grouping.
I agree that all of the resistors approach Johnson noise level at higher frequencies. For 100 ohm resistors up to 1 W, both SMD (Fig 6) and through hole (Fig 7), the lowest noise resistors approach Johnson noise level in the neighborhood of 100 Hz. The poorer group are all thick film with one bad thin film in the bunch. They get to Johnson noise level at higher frequencies.
At 1 kOhms and 10 kOhms, the best resistors reach Johnson noise level at similar frequency to their 100 Ohm cousins. Also, the higher power resistors reach Johnson noise at higher frequencies. The highest power group tested was all thick film design which is expected to have higher noise.
Note that in Figure 7, the author tested a bulk film resistor, Vishay S102K, which of course had the lowest noise.
I guess my point is that resistor manufacturers are continuing to improve. There are many resistors with lower than -40 dB noise and some quite a bit better. With the exception of the bulk film resistor and some parts from small makers that I can't find information on, everything here is from an industrial or military spec resistor. Many cost the same as resistors we are using now. Also, this study was from 2009. In the intervening 8 years, some of these resistors have become obsolete and replaced with better performing models.
If you can build your project with lower noise components for the same money, why not do it?
Last edited:
I think you are coming round to the selection idea of:
Exclude any badly made metal based types and all the better metals approach Johnson + 1/f noise. Done.
Now back to the series/parallel melfs and why they might be better than a single resistor.
There are some articles explaining this and it has been discussed a few times in this Forum.
There seems to be no evidence that shows that the series voltage equalising rule gets broken because of the extra joints.
The "test" report I saw, agian discussed in the Forum, showed a conventional Wheatstone bridge with a string of 10:1 on the REF side and a 10x value : 1 on the DUT side.
This allowed the extra distortion of the 10x value to show up and it was extremely small.
I don't recall any disagreement during those discussions that this voltage equalising rule was the solution to minimising the resistor distortions.
There was a new amplifier assembled and tested by one of our Members. It was based on a Blameless but distorted badly enough that the Member came asking for help. It turned out that one resistor (NFB upper?) in the build was approaching it's rated voltage at max power out. A series pair solved the distortion problem.
Exclude any badly made metal based types and all the better metals approach Johnson + 1/f noise. Done.
Now back to the series/parallel melfs and why they might be better than a single resistor.
There are some articles explaining this and it has been discussed a few times in this Forum.
There seems to be no evidence that shows that the series voltage equalising rule gets broken because of the extra joints.
The "test" report I saw, agian discussed in the Forum, showed a conventional Wheatstone bridge with a string of 10:1 on the REF side and a 10x value : 1 on the DUT side.
This allowed the extra distortion of the 10x value to show up and it was extremely small.
I don't recall any disagreement during those discussions that this voltage equalising rule was the solution to minimising the resistor distortions.
There was a new amplifier assembled and tested by one of our Members. It was based on a Blameless but distorted badly enough that the Member came asking for help. It turned out that one resistor (NFB upper?) in the build was approaching it's rated voltage at max power out. A series pair solved the distortion problem.
Last edited:
I agree that series and series/parallel resistor designs are an excellent approach to reduce THD+N. Especially useful in power resistor applications. As it happens, this amp has a great opportunity to try this idea. R3 is in the output signal path with a standard value of 0.47R and Evo A of 0.33R. At 0.47R it dissipates in the about 2.6W at full power.
Normally, we use a Caddock MP930 in this location with a heatsink. The MP930 is a good resistor with pretty good noise for a thick film design and the low value of the resistor should keep the noise fairly low. However, it should be possible to reduce noise further with a series/parallel metal film resistor. Since this resistor sees the whole music signal, it has the potential to be both a measurable and an audible improvement.
Dario mentioned the idea of using the MMA MELF resistors in a series/parallel combination. In the case of R3, I would choose 4 MMB0207 MELF resistors in series/parallel. The MMB is rated at 1W so we should have 4W combined. I'm not sure if it would get too hot if mounted on a small pcb as in the link the George referenced, but it's worth a try.
One last thought on selecting resistors. Johnson+1/f noise is a good starting place for selecting resistors, but there are other measures and more yet to understand. For example, in the Low Distortion Oscillator Design thesis by Samuel Groner (link below), Groner provides the best discussion that I have found of the power coefficient or thermal self-modulation of resistors. Components, both passive and active, are rarely ideal or simply defined. As we continue to improve our projects, I like to keep my eyes and ears open to new possibilities.
http://www.nanovolt.ch/resources/low_distortion_oscillators/pdf/low_distortion_oscillator_design.pdf
Jac
Wow! All of that and still having C9. I was expecting a bigger change with C9, more than the opamp. You are making me anxious to get busy with the soldering iron.
Jac
Wow! All of that and still having C9. I was expecting a bigger change with C9, more than the opamp. You are making me anxious to get busy with the soldering iron.
Jac
Hi Jac,
The LM318/Ada4627 opamp is the heart of the amplifier. It's linearitygain/bandwidth have a first order effect on determining the performance of the whole amp. I don't think anyone who does this mod will be disappointed. The only thing to keep in mind is that, like any amp, it does lose some character as it gets more neutral, and therefore gets more demanding of associated equipment.
Dan
I appreciate your assistance and encouragement George!
As for doing further mods, I like to give my ears plenty of time to get used to a sound. Maybe I'll get some time this weekend to do more.
I'd be interested to see the discussion of class A biasing. My only vague recollection is that people liked the LF01 which is a class A biased LM318 + buffer. I can't think the buffer makes that much difference driving a 47K load.
"I like to give my ears plenty of time to get used to a sound"
I could not agree more! Take your time.
About class A biasing: here you are!
http://www.diyaudio.com/forums/solid-state/3044-why-class-so-popular-3.html#post5090010
Ciao! G
I could not agree more! Take your time.
About class A biasing: here you are!
http://www.diyaudio.com/forums/solid-state/3044-why-class-so-popular-3.html#post5090010
Ciao! G
I am butting in. Dario, it is great to see what you have done to further the design of My_Ref. Reading the last few pages it looks like some advancements have been identified and built for testing. I need to look at the latest circuit diagram, not sure if items like C9 are the same as the last circuits I have seen.
I was in Rome this week when the news about Mauro reached me. I had exchanged emails with him trying to see if our schedules would allow a meeting. It was a shock Monday to see this will never occur.
When I return home a Evo Rev A is going back into my system. It was changed to allow using balanced connections. I really prefer balanced cabling.
I have methods to allow converting to balanced. Easiest is input transformers.
Plan on building another My_Ref. Whether a version 10, stuffing my last Evo board, or modifying one of the many I have built.
The design changes and component selection was always something that I tried to advance over a decade ago. Built one of the original gold plated My_Ref with all S102 resistors and other high spec components. It helps to be in Houston where Texas Components dumped surplus to the electronics shops.
From an honest opinion, does the version 10 deliver better sonics than an Evo Rev A Full?
I have an Evo Full that I can modify, a blank board and Galaxy chassis, and a Evo Rev A that I can add components to to make a full. But if the eight years of development since Rev A have surpassed this circuit It is time to move on.
I was in Rome this week when the news about Mauro reached me. I had exchanged emails with him trying to see if our schedules would allow a meeting. It was a shock Monday to see this will never occur.
When I return home a Evo Rev A is going back into my system. It was changed to allow using balanced connections. I really prefer balanced cabling.
I have methods to allow converting to balanced. Easiest is input transformers.
Plan on building another My_Ref. Whether a version 10, stuffing my last Evo board, or modifying one of the many I have built.
The design changes and component selection was always something that I tried to advance over a decade ago. Built one of the original gold plated My_Ref with all S102 resistors and other high spec components. It helps to be in Houston where Texas Components dumped surplus to the electronics shops.
From an honest opinion, does the version 10 deliver better sonics than an Evo Rev A Full?
I have an Evo Full that I can modify, a blank board and Galaxy chassis, and a Evo Rev A that I can add components to to make a full. But if the eight years of development since Rev A have surpassed this circuit It is time to move on.
Thanks
In the My_Evo there's no C9, it's function is done by the DC-Servo
Absolutely...
What's version 10? Are you referring to my My_Ref variant (Fremen Edition) or what?
If so probably JosephK can give a better answer since he owns both, personally I've tried the My_Evo for a week or so several years ago, with the previous BOM.
From what I can remember the My_Evo was in general a bit superior but regarding detail and harmonic content the FE was better, IMHO the ceramics on the My_Evo were the culprits.
The My_Evo circuit is pretty similar to the Fremen Edition, from a circuitry perspective the My_Evo is more advanced.
In fact the Fremen Edition is a re-implementation of My_Evo goals without infringing Mauro's IP.
The FE has no DC-servo, a different power supply and no support for the double pump.
The FE has the advantage of an higly (sound) optimized BOM.
JosephK experimental opamp swap can be applied to both amps, in the FE you can avoid C9, in the My_Evo the DC-Servo and its horrible big ceramic.
Thank you Dario,
Your work is impressive. The circuit is like My_Ref C? If not, is there a page for n this thread that has been he diagram?
Couple items, I started added some 22 UFD, 50v Panasonic caps right to the power pins of the 3886. Connect the center to the nearest power ground.
The other item is the output resistor. Used a Dale 5W Current Sense resistor in a value of 0.536 ohms in the My_Ref. These are low inductance and non-magnetic. A little off in value.
For the Evo ref A used two in parallel and a 0.05 in series to get the 0.33.
The Dale resistors were dumped by a company. Saw them at Ace and EPO in Houston and Tanner in Dallas. They are helictical construction.
Your work is impressive. The circuit is like My_Ref C? If not, is there a page for n this thread that has been he diagram?
Couple items, I started added some 22 UFD, 50v Panasonic caps right to the power pins of the 3886. Connect the center to the nearest power ground.
The other item is the output resistor. Used a Dale 5W Current Sense resistor in a value of 0.536 ohms in the My_Ref. These are low inductance and non-magnetic. A little off in value.
For the Evo ref A used two in parallel and a 0.05 in series to get the 0.33.
The Dale resistors were dumped by a company. Saw them at Ace and EPO in Houston and Tanner in Dallas. They are helictical construction.
Hi Panelhead,
Happy to see another experimenter on the thread. The schematic is on the first page of this thread. Look about 3 lines down and see the link for the Google Drive folder. In that folder, Dario has placed BOMs for each level of FE and a schematic.
I am interested in your Dale for R3. You may have noticed the posts in the last few pages about using series/parallel combination of low noise metal film resistors. George suggested we look at that idea discussed in another format. I plan on trying the idea using 4 MMB0207 MELF resistors on a small perf board to get the same packaging as the Caddock. The MMB is already a pretty good resistor and the benefits of series/parallel combination further improve noise and distortion.
Happy to see another experimenter on the thread. The schematic is on the first page of this thread. Look about 3 lines down and see the link for the Google Drive folder. In that folder, Dario has placed BOMs for each level of FE and a schematic.
I am interested in your Dale for R3. You may have noticed the posts in the last few pages about using series/parallel combination of low noise metal film resistors. George suggested we look at that idea discussed in another format. I plan on trying the idea using 4 MMB0207 MELF resistors on a small perf board to get the same packaging as the Caddock. The MMB is already a pretty good resistor and the benefits of series/parallel combination further improve noise and distortion.
@Panelhead:
Upon George's suggestion and some tests, lately we've completely removed the DC servo from our own my_Evos!
As it turns out, indeed it sounds better without it.
(N.B.: of course we are also using JFET-input OpAmps -in my case ADA4627- instead of the LM318, otherwise the servo could not be removed).
Thus, currently our updated my_Evos are really a lot similar to Dario's FEs (with George's mods).
The main (and almost only really significant) difference which remains between the Evo and the FE is the possibility of doubling the current pump, which does exists in the Evo but not in the FE.
In our experiences, the "full-Evo" (with double c.pump) does indeed have some notable advantages -also sound-wise- WRT a single-pump version. At least when paired with speakers which shows lowerish impedances. With "easy load" speakers that have a pretty flat impedance curve, with modest phase shift and which never goes much below 8 Ohm, perhaps the difference might be less evident.
Thus, if you have an Evo board, I'd use that one. In "full" configuration (4*3886, etc), with all George mods. In that configuration and using the right BOM, it's SQ will rival that of the very best amps on the market, no matter the price!
Upon George's suggestion and some tests, lately we've completely removed the DC servo from our own my_Evos!
As it turns out, indeed it sounds better without it.
(N.B.: of course we are also using JFET-input OpAmps -in my case ADA4627- instead of the LM318, otherwise the servo could not be removed).
Thus, currently our updated my_Evos are really a lot similar to Dario's FEs (with George's mods).
The main (and almost only really significant) difference which remains between the Evo and the FE is the possibility of doubling the current pump, which does exists in the Evo but not in the FE.
In our experiences, the "full-Evo" (with double c.pump) does indeed have some notable advantages -also sound-wise- WRT a single-pump version. At least when paired with speakers which shows lowerish impedances. With "easy load" speakers that have a pretty flat impedance curve, with modest phase shift and which never goes much below 8 Ohm, perhaps the difference might be less evident.
Thus, if you have an Evo board, I'd use that one. In "full" configuration (4*3886, etc), with all George mods. In that configuration and using the right BOM, it's SQ will rival that of the very best amps on the market, no matter the price!
Last edited:
Dear all,
Very happy to see also George joining us, -- again
And Paolo has just taken the words out of my mouth, I could not say it better
And just a funny note for Dario:
Hehe, this was an idea of Mauro that could not get through 'my potential barrier'..
Right at the moment of the rev-A modification executed by me on the EVO amp, 'that horrible ceramic cap' got already swapped out for a pair of PPS film caps..
Still, suspending the dc-servo circuit had brought a notable clean up of the sound, recently.
Ciao, George
Very happy to see also George joining us, -- again
And Paolo has just taken the words out of my mouth, I could not say it better
And just a funny note for Dario:
Hehe, this was an idea of Mauro that could not get through 'my potential barrier'..
Right at the moment of the rev-A modification executed by me on the EVO amp, 'that horrible ceramic cap' got already swapped out for a pair of PPS film caps..
Still, suspending the dc-servo circuit had brought a notable clean up of the sound, recently.
Ciao, George
George,
Is it the resistor that You mention? LVR-05 or LVR-010?
http://www.vishay.com/docs/30206/lvr.pdf
The higher resistance values seem to have a good temp. coefficient; (50-100 ppm)
it looks a good choice.
I am using something similar: the ATE series 5CS wire-wound resistors, I think it is the one, or similar to the one that Mauro had been using for the rev-A mod.
Ciao, George
Is it the resistor that You mention? LVR-05 or LVR-010?
http://www.vishay.com/docs/30206/lvr.pdf
The higher resistance values seem to have a good temp. coefficient; (50-100 ppm)
it looks a good choice.
I am using something similar: the ATE series 5CS wire-wound resistors, I think it is the one, or similar to the one that Mauro had been using for the rev-A mod.
Ciao, George
Yes LVR-5
I think these were custom order and over built. Found them at three surplus parts stores. The value of 0.536 is the highest I found. Most are 0.03 - 0.1 ohms.
I am learning a lot already. Swap opamp, remove servo, I look forward to discussing these.
Maybe I can contribute a little to this.
I think these were custom order and over built. Found them at three surplus parts stores. The value of 0.536 is the highest I found. Most are 0.03 - 0.1 ohms.
I am learning a lot already. Swap opamp, remove servo, I look forward to discussing these.
Maybe I can contribute a little to this.
Thanks for the link George. Interesting reading. Anyway I've tried it, with 2.7mA current regulator diodes pulling current to the negative rail.
Very obviously sounds warmer, which is welcome. i'd be curious if this is reflected in distortion measurements, if you ever have the chance!
C9 next.
Can you share the parts you used and a rough schematic of the class A biasing?
Thanks
Jac
Sure Jac.
I used a CRD, which is basically a self-biased JFET current source.
http://www.mouser.co.uk/ProductDetail/Semitec/E-272/?qs=wgO0AD0o1vtDDfE6Oj1ALA==
It's connected between the output of the opamp (in my case ADA4627) and the negative power supply. This one is rated at 2.7mA, plenty enough to ensure class A operation at all levels.
Dan
I used a CRD, which is basically a self-biased JFET current source.
http://www.mouser.co.uk/ProductDetail/Semitec/E-272/?qs=wgO0AD0o1vtDDfE6Oj1ALA==
It's connected between the output of the opamp (in my case ADA4627) and the negative power supply. This one is rated at 2.7mA, plenty enough to ensure class A operation at all levels.
Dan
Last edited: