I'd suggest that if there is an argument you don't think is worthwhile, you don't participate. It takes at least two to derail a thread.
And I have little interest in the subject beyond making the odd passing comment, because, although I believe it's an established fact that our ears can deceive, proselytising this fact isn't any fun, least of all among audiophiles.
I will, however, continue to post what I please, so long as it is within the forum rules, and 99.9% of this you will find completely uncontentious. If you ignore the other 0.1%, it probably won't go anywhere.
Anyway, my last word, as I don't want to derail things further.
Hi Tom,
32 mils trace witdht, 70u thickness, lenght as indicated (1mm and 5.5mm)
No, it's my variation: My_Ref Fremen Edition
There are several versions:
- the original stereo layout by Mauro Penasa (see Mauro's website for schematic and gerbers)
- the monoblock layout by Russ White of Twisted Pear Audio, it's the most used and out of production (released for personal use)
- LinuxGuru monoblock layouts (somewhat an unofficial evolution from TP boards)
- Nimo e-store unauthorized stereo clone
- XCalibre
- My_Ref Fremen Edition (completely redesigned PCB, PS, groundplanes, different grounding, SMD parts, etc.)
- LinuxGuru double pump variation (yet to be released)
http://www.diyaudio.com/forums/chip-amps/197120-my_ref-fremen-edition-need-help-pcb-evaluation.html
http://www.diyaudio.com/forums/chip-amps/207390-my_ref-fremen-edition-beta-build-fine-tuning.html
http://www.diyaudio.com/forums/chip-amps/216915-my_ref-fremen-edition-rc-build-thread.html
The FE is in the final development phase (release candidate) but suggestions and (constructive) comments are always welcome and appreciated.
The ony currently developed versions/variations are the FE and Siva's ones
Last edited:
I'll add that the last Twisted Pear Audio (v1.2) and my v1.3 and v1.4 PCBs are closely related, with nearly identical schematics and parts designators. My boards are physically larger, because one of the reasons for the respin was to allow physically larger components to fit comfortably, e.g. 35mm diameter PSU caps (C3,C8) vs. 30mm earlier, and 16mm diameter caps at C1,C2 vs. 12mm earlier. There are also numerous changes to allow component flexibility, while allowing components from earlier BoMs to be used interchangeably wherever possible. There are some routing improvements, most visible in V1.4 around C4, C7 and the PGND nets, and better shielding/ground plane for the small-signal nets.
The double-Howland MyRef-X2 is an ab-initio redesign and fresh layout, and it is still in the process of prototype validation. The only change in specs is that it can drive 4-ohm loads also, with the same nominal 32V rails for a power rating of 80 Wrms per channel. Specs for 8-ohm loads remain unchanged.
The double-Howland MyRef-X2 is an ab-initio redesign and fresh layout, and it is still in the process of prototype validation. The only change in specs is that it can drive 4-ohm loads also, with the same nominal 32V rails for a power rating of 80 Wrms per channel. Specs for 8-ohm loads remain unchanged.
In simulation, with unofficial LM1875 models instead of LM3886, there is indeed a small improvement in THD20 into 8-ohm loads compared to the single-pump case. I'm not sure if it will be audible - the standard MyRef was already very good.
Last edited:
I want to be discrete and polite, but did you find any of DW's comments in this section of the discussion of "amp sweetness" applicable to the X2, or if implemented, an entirely new PCB (non-MyRef) would be required?
I'm glad you liked the link. I thought it was interesting, and relevant. And there are many other good links, at that link, and also at the homepage of that site. (They use many, many animations, to illustrate circuit operations.)
Low impedance of power and ground connections at chips is so important that I am spending quite a lot of time thinking about it.
(The following are general ideas and suggestions, for future designs, not necessarily suggestions for the particular layout(s) already being discussed in this thread.)
We need some way to get a Terry Given-style capacitor bank board right up to each power pin. I think maybe we need to throw away most of the conventional layout ideas and make a layout based around a cap bank for each power rail, where the cap bank board IS the power and ground rails.
Per Terry's posts (see below for link), I would use a 1mm-thick FR4 double-sided PCB. One side is a power plane and one side is ground plane. Both sides are left as all solid copper except for holes for cap leads, with a small amount of copper removed from around the edge of one hole per cap, on cap side of board. Cap leads are cut to roughly 12 mm, bent flat against copper, and soldered. A 10 x 10 array of capacitors could be used for each rail; perhaps 1000 uF for each cap.
The chipamp would probably need to be at the midpoint of one side of a square-ish cap array, with one array for each rail. So the amp would need to be between two cap arrays. And each array would need to be fed at the midpoint of the opposite side of the square, relative to the chipamp. Looks like caps and heatsink would need to be on opposite sides of the board. (Alternatively, maybe the whole thing could be "folded" in the middle, where the two arrays meet, so that two boards were used, that were parallel to each other. Then the rectifier connections would both be at the same end. But the chipamp would need to connect to both boards.)
By using that type of cap-array power/ground-plane board, we could have a total of less than 1 nH (maybe 0.5 nH) of impedance presented between each power rail and ground, i.e. between each chipamp power pin and any nearby ground plane location (and still less than 0.1 Ohms @ 2MHz, rising to 1 Ohm at 90MHz), with no resonances to at least 200 Mhz (the limit of the network analyzer that was used). So no other bypass or decoupling caps would be necessary. And we would have ground plane available all around.
Still have to figure out the best way to mount the other passive components that need to be near the chipamp, without breaking up the power and ground planes. Maybe a daughterboard, mounted vertically, parallel to and in front of chipamp. Maybe some of the chipamp pins could be bent up to go directly to the daughterboard.
OR, the daughterboard could be mounted parallel to the main board, with short standoffs, right under the chipamp, so that chipamp pins could go to the daughterboard instead of the power/ground board, as needed, while other pins could pass through to the main pwr/gnd board, or both. That might actually work pretty well.
I collected all of the links to Terry Given's posts about the capacitor bank, in a post at:
http://www.diyaudio.com/forums/chip-amps/224914-lm3886-component-selection-3.html#post3282640
Low impedance of power and ground connections at chips is so important that I am spending quite a lot of time thinking about it.
(The following are general ideas and suggestions, for future designs, not necessarily suggestions for the particular layout(s) already being discussed in this thread.)
We need some way to get a Terry Given-style capacitor bank board right up to each power pin. I think maybe we need to throw away most of the conventional layout ideas and make a layout based around a cap bank for each power rail, where the cap bank board IS the power and ground rails.
Per Terry's posts (see below for link), I would use a 1mm-thick FR4 double-sided PCB. One side is a power plane and one side is ground plane. Both sides are left as all solid copper except for holes for cap leads, with a small amount of copper removed from around the edge of one hole per cap, on cap side of board. Cap leads are cut to roughly 12 mm, bent flat against copper, and soldered. A 10 x 10 array of capacitors could be used for each rail; perhaps 1000 uF for each cap.
The chipamp would probably need to be at the midpoint of one side of a square-ish cap array, with one array for each rail. So the amp would need to be between two cap arrays. And each array would need to be fed at the midpoint of the opposite side of the square, relative to the chipamp. Looks like caps and heatsink would need to be on opposite sides of the board. (Alternatively, maybe the whole thing could be "folded" in the middle, where the two arrays meet, so that two boards were used, that were parallel to each other. Then the rectifier connections would both be at the same end. But the chipamp would need to connect to both boards.)
By using that type of cap-array power/ground-plane board, we could have a total of less than 1 nH (maybe 0.5 nH) of impedance presented between each power rail and ground, i.e. between each chipamp power pin and any nearby ground plane location (and still less than 0.1 Ohms @ 2MHz, rising to 1 Ohm at 90MHz), with no resonances to at least 200 Mhz (the limit of the network analyzer that was used). So no other bypass or decoupling caps would be necessary. And we would have ground plane available all around.
Still have to figure out the best way to mount the other passive components that need to be near the chipamp, without breaking up the power and ground planes. Maybe a daughterboard, mounted vertically, parallel to and in front of chipamp. Maybe some of the chipamp pins could be bent up to go directly to the daughterboard.
OR, the daughterboard could be mounted parallel to the main board, with short standoffs, right under the chipamp, so that chipamp pins could go to the daughterboard instead of the power/ground board, as needed, while other pins could pass through to the main pwr/gnd board, or both. That might actually work pretty well.
I collected all of the links to Terry Given's posts about the capacitor bank, in a post at:
http://www.diyaudio.com/forums/chip-amps/224914-lm3886-component-selection-3.html#post3282640
Last edited:
Recently I have been looking at ground noise and it's effect on audio. Quite interestingly, when the noise has a complicated set of frequencies, noise is not as audible. Once you reduce the noise to limited frequency content, then it can become more annoying. However, with less frequency content in the noise, sound is cleaner and more focused revealing more detail. I probably need a better scope to look at the noise, but it was interesting how ground noise feeds back into the circuit. Since ground is really a theoretical reference in circuit performance, we really need to look at both the supply and return currents and the impedance they see. There is no easy formula. Each layout and circuit need to be individually considered. We talked about swapping positions of resistor and caps somewhere, and that is a good example of handling of return currents.
Thank you for the interesting information and ideas, soongc.
Possibly, it should "always" be possible to create ground reference points that contain no CONDUCTED current-induced voltage noise from other parts of the same circuit or system, just by making sure that the ground reference shares NO length of conductor with any other ground return conductor, on its way to the main or star ground point. That should avoid "ground bounce" due to V = L di/dt and V = ir from currents from other parts of the circuitry, which would be induced at any ground point that shared any length of conductor with other ground currents.
The worst-case potential for that is probably at the signal input ground reference point, at the ground end of the input resistor that connects to the input pin of the first active amplifiying device. Any time-varying "ground" voltage there would be arithmetically summed with the input signal. Run a separate dedicated conductor from signal input area to star ground.
However, a ground reference with its own dedicated separate conductor to the star ground point would still be able to have radiated/coupled noise induced in it. However, in that case, maybe using a shielded cable, instead of a pcb trace or open wire, might be helpful.
(I hope you don't mind if I more-or-less just regurgitate, for the record, my usual list of techniques for minimizing noise and other problems
Choice of the star ground point is usually important, along with making sure that rectifier charging-pulse currents do not make it to the star point, but rather are kept in a small loop through only (and not all of) the reservoir caps.
Use good decoupling to keep power and ground rails cleaner. Use multiple parallel caps extremely close to exact point of load, for active devices, usually grounded to load ground. At slightly farther distances from power pin, start adding larger caps. Decoupling layout should be highest priority, in an amplifier, after PSU, since the output signal IS the current, which comes directly from the PSU caps and the decoupling caps.
(One of my next amplifier builds will also include "star power", paired with star ground returns, with minimized enclosed loop area (unless I use power and ground planes).
Also, "Don't make antennas". Any natural pair of conductors, i.e. every conductive loop (and current only flows in loops), must have its enclosed geometric area eliminated, as much as possible. The worst offenders as "transmitters" are AC input and transformer input and secondary wire pairs, and the pairs on both sides of rectifier bridge. Also, speaker pairs. ALL must be tightly twisted together (or run as shielded twisted pair with shield grounded only to chassis, only at one end). If pcb traces are used instead of wires, use 2-sided board and run pairs exactly overlapping each other, on opposite sides of board. Make planes on opposite sides, if possible. For 1-sided board, run pairs as traces that are never separated, with minimum gap between them.
Worst "receiving antenna" case is signal and signal ground. Should be run as shielded twisted pair with shield grounded only to chassis, only at one end, or, run as tightly-twisted wire pair. Must stay twisted ALL the way from input jack to the input pin of the first active amplifier component, with no contact with chassis or ground. Pair can separate right at input pin shunt resistor, connecting to each end. Separate conductor should run from input shunt resistor's "ground" end to star ground. On a 2-sided PCB, always use a dedicated separate signal-ground ground plane, overlapping ALL components connected to input signal or ground. Run separate dedicated conductor from signal ground plane to star ground. On 1-sided PCB, keep signal and signal ground traces ALWAYS together, with minimum gap between them. Flood adjacent areas with dedicated signal ground copper, including under ALL components that connect to signal input network, and both sides of signal trace if possible.
All other pairs and loops should be made to have absolute-minimum enclosed loop area.
Also, low-pass or other RF filters or RF mitigation should be installed on each input, on the output, and at the power rail inputs. ("Everything is an input, for RF.")
Cheers,
Tom
There are really a few separate issues. For example, lots of focus is on the noise level at idle. Now, when you measure this, how can one distinguish whether it's noise in the ground or noise in the power rail? Then there is the issue of the impedance of the power; I think this is very controversial because other issues like how the return current path flows. It will flow first into whatever cap is connected to it, so that sort of feeds back into the circuit depending one how your circuit is coupled with the ground. These are very complicated interactions. No easy way of analyzing all this. I just recently started to get a better feeling how to handle this in a voltage source amp, but when you get to current source amps and amps like the MyRef which is a combination of both, this is even more complicated.
One way to TRY to analyze it is to make a model of the circuit, in Spice (I use LT-Spice, which is a free download from linear.com, but is very highly regarded). I usually add the parasitics to the component models, first. There are even frequency and temperature dependent models for electrolytic caps available, from a Java applet at the Cornell Dubilier site. Or, you can just enter numbers (or equations) for ESR, ESL, and even EPR if desired, by right-clicking on each cap.
Then (if desired) I add the inductance and resistance for each interconnection, by inserting inductors and resistors in the schematic, in LT-Spice.
I now usually parameterize the length of each connection and enter the value of each parasitic as an equation based on the length variable's name that I chose for that trace or wire and the (also-parameterized) inductance or resistance per unit of length, so I can later change any or all of the parameters, easily, and in one (or at least fewer) places, and have the program automatically calculate everything, at runtime.
That (parameterization) also provides the ability to SWEEP any parameter (or up to three or four nested sweeps, actually), at any later time, and automatically initiate and run a whole series of simulation runs (one per sweep step, for each sweep), with all of the results plotted simultaneously, if desired. One can also specify almost any kind of automatic calculations for post-analysis, based on any quantities in the circuit or the simulation results.
When I first started trying to model PCB and wire parasitics, it was in order to try to see the possible effects of letting an amplifier's signal input ground reference share a conductor (back to the star ground) with different types of ground returns, such as decoupling cap ground returns, and speaker ground returns, of varying lengths, under various conditions.
It's really not too difficult, although it can sometimes become tedious, for large and complex circuits. However, it is obviously just a rough approximation. And I have never actually tried to formally validate and verify the results of simulating with conductor parasitics. (There are probably already technical papers or articles about using Spice for that, though.)
I wonder if there is any freeware out there that will take a simple one-sided (and maybe also two-sided) PCB layout drawing, and some specs about the materials, and calculate the models that would be needed for a Spice simulation to include the PCB parasitics. Seems like it might be not TOO terribly difficult to create such an app, to at least get a significantly-better approximation than just using 1 nH of self-inductance per mm and 1/25.4 mOhm per mm.
Last edited:
I have a few steps in mind in which to approach this, and currently only at the initial stage. First is to rely on engineering intuition to review the layout and the circuit together, and figure out what to try. The good thing now is that with a PCB milling machine, turn around time for a new PCB is about a day. Listening tests are conducted as well as looking at the scope and spectrum analyzer. Generally I make changes in one channel at a time to compare against the other. Improvement seems to be associated with darker background, finer details, sounding lower in volume, slightly more spaciousness, overall more dynamic and emotionally moving. I think until I get to a stage where Engineering intuition becomes less productive, and that I still have more desire for improvements, then I will go to the analysis stage.
The analysis stage is more comprehensive, takes more time, and also involves lots of effort to correlate analysis data with listening perceptions. This is necessary when more precise tuning is required. This stage requires model verification of each component as well as PCB parasitics. I don't think that we will ever spend the effort to perfect this in audio. Perhaps the CPU industry uses something that does this better which I can see reasonable for such investment. But it is reasonable to get the simulation to more correctly model the asymmetry nature of components and power supply so that you can properly now how the currents flow and fine tune that.
I generally use SoundEasy for concept exploration because it is good for quick work. Packages like LT Spice, Intusoft, OrCad, Mentor, Dazix, and a whole bunch of others that have merged throughout the years each have very interesting features. I don't think I have stuck with one long enough to see anything special about them mainly because nobody gives you well comprehensive answers when you start asking detailed questions what is simulated and how detailed are models verified.
The analysis stage is more comprehensive, takes more time, and also involves lots of effort to correlate analysis data with listening perceptions. This is necessary when more precise tuning is required. This stage requires model verification of each component as well as PCB parasitics. I don't think that we will ever spend the effort to perfect this in audio. Perhaps the CPU industry uses something that does this better which I can see reasonable for such investment. But it is reasonable to get the simulation to more correctly model the asymmetry nature of components and power supply so that you can properly now how the currents flow and fine tune that.
I generally use SoundEasy for concept exploration because it is good for quick work. Packages like LT Spice, Intusoft, OrCad, Mentor, Dazix, and a whole bunch of others that have merged throughout the years each have very interesting features. I don't think I have stuck with one long enough to see anything special about them mainly because nobody gives you well comprehensive answers when you start asking detailed questions what is simulated and how detailed are models verified.
Last edited:
Link to a fully-discrete Class-A folded Kaneda-style bipolar single-opamp in DIP8:
http://www.diyaudio.com/forums/analog-line-level/10243-kaneda-preamp-11.html#post3348288
The first pair of prototypes has been tested in a MyRef Rev E premium, and has sonics which are comparable to my earlier LF01 hybrid/discrete opamp module - transparent, open and smooth. I expect further improvements after experimentation with better active devices, especially the LTP.
http://www.diyaudio.com/forums/analog-line-level/10243-kaneda-preamp-11.html#post3348288
The first pair of prototypes has been tested in a MyRef Rev E premium, and has sonics which are comparable to my earlier LF01 hybrid/discrete opamp module - transparent, open and smooth. I expect further improvements after experimentation with better active devices, especially the LTP.
My_Ref Fremen Edition
For those interested the GB for the My_Ref Fremen Edition final boards has started.
http://www.diyaudio.com/forums/group-buys/229358-my_ref-fremen-edition-interest-group-buy.html
For those interested the GB for the My_Ref Fremen Edition final boards has started.
http://www.diyaudio.com/forums/group-buys/229358-my_ref-fremen-edition-interest-group-buy.html
It has been a whirlwind two days looking at amps and learning some - much of which I probably can't remember. However, after some thought and being a control freak to some degree, I decided I want to build a LM3886 amp for some 30W speakers I'll build soon.
I am now looking for an opinion or two regarding the normal LM3886 amp build, vs (I hate that term but alas) the MyRefC V1.3 build.
My application is as follows:
- 30W Full Range speakers (first build) - Visaton FRS8-8 drivers since they are freely available here.
- Crappy room layout but can't be changed
- Will be used in conjunction with my laptop as audio source
- Near field application
- Speakers used for audio reference whilst editing video
- I'm no audiophile
- Likely listen to some music as well, watch a movie.
Why build? It is fun. I have been meaning to build some audio project. I can't get myself to purchase those PC accessory speakers no matter the reviews - tried about 5 times over the last year.
What would you recommend, the normal LM3886 (like BrainGT has on his website) amp layout as from the manufacturer with maybe some small improvements in parts as mentioned in another MyRefC thread, or the MyRefC V1.3 build?
The reason I ask is I found some spare boards someone is selling on ebay. Pretty much the only reason I am considering it - and if for a modest cost extra it makes a world of difference, then I'm a go. Can I get an idea of the cost difference? The V1.3 boards aren't that much more that the standard boards (off ebay). It is the extra components involved that bugs me.
I read some of the New threads and group buy threads but it really is above my head. From the looks of the board alone it seems to be quite a bit more expensive.
My sincerest apologies for this beginner question, I am getting lost in all the threads of all the builds since most of the discussions are about layouts and using this cap over that cap which gets me lost.
I am now looking for an opinion or two regarding the normal LM3886 amp build, vs (I hate that term but alas) the MyRefC V1.3 build.
My application is as follows:
- 30W Full Range speakers (first build) - Visaton FRS8-8 drivers since they are freely available here.
- Crappy room layout but can't be changed
- Will be used in conjunction with my laptop as audio source
- Near field application
- Speakers used for audio reference whilst editing video
- I'm no audiophile
- Likely listen to some music as well, watch a movie.
Why build? It is fun. I have been meaning to build some audio project. I can't get myself to purchase those PC accessory speakers no matter the reviews - tried about 5 times over the last year.
What would you recommend, the normal LM3886 (like BrainGT has on his website) amp layout as from the manufacturer with maybe some small improvements in parts as mentioned in another MyRefC thread, or the MyRefC V1.3 build?
The reason I ask is I found some spare boards someone is selling on ebay. Pretty much the only reason I am considering it - and if for a modest cost extra it makes a world of difference, then I'm a go. Can I get an idea of the cost difference? The V1.3 boards aren't that much more that the standard boards (off ebay). It is the extra components involved that bugs me.
I read some of the New threads and group buy threads but it really is above my head. From the looks of the board alone it seems to be quite a bit more expensive.
My sincerest apologies for this beginner question, I am getting lost in all the threads of all the builds since most of the discussions are about layouts and using this cap over that cap which gets me lost.
A plain Gainclone is very different animal from a My_Ref.
The Gainclone focus is on layout (as small as possible) and parts quality relying on the power opamp performance.
The My_Ref uses the same opamp used in Gainclones as a current pump in a more complex (and much better sounding) approach.
My_Ref can be influenced from parts selection but even with plain industrial parts it will sound better than a Gainclone, IMHO.
I think you should identify first your main goal: audio performance? price?
Most of the times those two are in conflict.
The Gainclone focus is on layout (as small as possible) and parts quality relying on the power opamp performance.
The My_Ref uses the same opamp used in Gainclones as a current pump in a more complex (and much better sounding) approach.
My_Ref can be influenced from parts selection but even with plain industrial parts it will sound better than a Gainclone, IMHO.
I think you should identify first your main goal: audio performance? price?
Most of the times those two are in conflict.