Which PCBs have you experienced deterioration when the local decoupling is installed?
This is a sure sign of poor layout/earthing/PSU wiring.
You need to keep yucky stuff from your Clean earth which is used ONLY for signal & feedback. They should join at ONE point.
The best way to design this is to consider what happens if your earth tracks have significant resistance & inductance ... cos they DO!
In comparison with other criteria, this effect would probably of least sonic effect. Generally would be masked by the loudspeakers. I would not have discovered it if other had not reported different sonics with different caps on the Twisted Pear MyRef RevC boards. It is amazing what you can learn when different people try different things.
There is no clean ground in reality! I discovered this in the lab doing conductive noise testing of a product trying to figure out what to do. Use a Class II connection, all is perfect, use a Class I connection and you see the noise. Probably the linear power supplies would be better. The trick is to creat different resistive paths to channel your ground current. You can also do this to control current throughout the circuits...
There is no clean ground in reality! I discovered this in the lab doing conductive noise testing of a product trying to figure out what to do. Use a Class II connection, all is perfect, use a Class I connection and you see the noise. Probably the linear power supplies would be better. The trick is to creat different resistive paths to channel your ground current. You can also do this to control current throughout the circuits...
Yes, totally integrated into an active speaker. Includes equalization and zobels to match the speaker. Most time was spent trying to optimize the integration for best fidelity possible.
The LM3886 is quite amazing when things are worked out. I had tried an early configuration in the MyRef which I build and shown in the original thread. I am a sucker for trying to make things small and perform good.
The LM3886 is quite amazing when things are worked out. I had tried an early configuration in the MyRef which I build and shown in the original thread. I am a sucker for trying to make things small and perform good.
Last edited:
The MyRef layouts actually use 4 rail traces to the 4 bypass capacitors, and 2 ground traces from each bypass pair back to star ground - there isn't really any other way to do it correctly. There's no small film cap from any rail to ground - only from rail-to-rail near the chipamp, and another near the opamp. This avoids contaminating either the chipamp ground or the opamp small-signal ground reference with fuzz. The rail traces are pretty wide, BTW.
The MiniRef does mostly the same thing, but integrates two channels (common rails) on the same board. It works and sounds fine, but the channel separation may not be as high as the MyRef, due to the shared rails and feedthrough from the rails to the other channel.
Cool - another idea could be to integrate it into a subwoofer box for a 2.1 setup. It's not a new idea, and there are zillions of such 2.1 products with a sub and 2 satellites, mainly for the computer, HTPC and personal music-player market. The sub could possibly be driven by a Class-D amp, and the satellites by Class-AB chipamps like a MyRef or MiniRef.
I was just listening to a Sonicgear iSteroid-1 tube-buffered 2.1 sub-woofer (with a 6N2 cathode follower driving fairly basic chipamps with 10W output to the sub-, and 2.5W each to the satellites) this week, and it sounded surprisingly good, given the fairly ordinary quality of the components. The same concept, with high-quality gainclones for the satellites, and Class-D for the sub-, could be giant killer.
Last edited:
The local rail-to-rail decoupling caps are very close to the chipamp and opamp, respectively - not more than a few mm in all cases.
The track lengths from the main reservoir/filter caps to the chipamp are not that important - they just need to be wide enough and enclose a small radiating loop area (I've seen the same layout technique in midrange Japanese consumer electronics amps from the '80s/'90s - Yamaha, Denon, Technics, etc.). They typically have a small (~4.7 to 10 uF) electrolytic from each rail to ground not too far from the chipamp or module. That ground is a separate wire or trace back to star ground, which is located between or close to the main electrolytics.
Actually one common effect of poor layout/decoupling is marginal instability with certain loads/levels .. all dependent on the thermal history.
So it is perfectly possible to hear differences in Double Blind Listening Tests due to poor layout/decoupling etc.
Did you notice this problem with other PCBs beside the Twisted Pear?
Of course how the PCB & PSU is wired up also makes a big difference.
linuxguru's scheme in #128 is a good one.
Last edited:
The MyRef PCB designed by Russ White has for the LM3886 a 220uF plus 100nF per rail and a 100nF rail to rail with a single trace to PSU ground. The LM318 has the same configuration only 100uF instead of 220uF caps.
Load return is connected to PSU ground which is also against TI recommendations, should be returned to the Bypass caps.
Maybe it explains why for some the TI recommended bypass values don't work as expected.
Last edited:
The best sounding solution on the Twisted Pear boards were increasing the cross rail caps and changing to snubbers. I have explained some of my experiments in the original thread.
The same on another chip amp I was working on, and limited space also limited my options. I will try some things when working on a stand alone amplifier. But I really which there were a way to look at real data from TI showing the use vs no use of decoupling caps, and the actual board they used for such test. I have seen enough about snubbers, and understand how they help, and consistent with my listening experience. The decoupling caps issue seems more important when you are feeding it ultrasonic signals. I am wondering it might also help in EM susceptibility. Perhaps the need will manifest itself when I get to that.
For serious critical music listening, I think 2.1 is not good. I would vote for full stereo subs.
Are you describing Soongsc's implementation?
or what should be done.
Soongsc's description that I re-quoted does not seen at all like what you have described.
These two posts seem to be suggesting that all your 3886 implemenatations from simple to MyRef have the same fault. They don't perform properly with local decoupling and further suggest that removing the local decoupling is the best solution for improved performance. Further, that adding some snubbing (R+C?) will improve still further.
Is this saying that all of your previous applies only to MyRef?
or do your first two posts still hold true?
I am not sure what you are asking AndrewT. I discovered certain possible improvement in power supply decoupling to improve sound in the MyRef which I also verified and use in chip amps. Just sharing my experience. If anyone had tried out the various options that I had tried, and maybe others, I certainly would like to hear about it.
One thing I try to avoid is getting into which is right or wrong issue unless I have data to show, which I do not. One thing I do is that I mostly try things if I at lease have a good technical reasoning behind it, if the result may be component value dependent, then I need measurements to help with finding that design point, which happened during design of a line power filter.
One thing I try to avoid is getting into which is right or wrong issue unless I have data to show, which I do not. One thing I do is that I mostly try things if I at lease have a good technical reasoning behind it, if the result may be component value dependent, then I need measurements to help with finding that design point, which happened during design of a line power filter.
Mauro didn't have 100nF per rail film bypass caps, only electrolytics and the rail-to-rail film cap. Russ White added the rail-to-ground film caps, but far from improving the audible sonics, it actually seemed to degrade it. The jury is still out on why that happens, but it could be due to ground contamination with audible fuzz.
I was aware of this option, but didn't want to risk degrading the
sonics of the proven Twisted Pair layout, so I stayed with routing
the speaker return to the PSU star ground in all 4 layouts - Rev C
v1.3, v1.4, MiniRef 1875 and MiniRef 3886. It hasn't hurt, but maybe
I'll try some alternatives in a future layout.
Linux,
do you remember that we discussed the routing of the main smoothing caps link on the OTHER side of the PCB?
So far you are the only person to implement a PSU Zero Volts link to Main Audio Ground by using a single trace that is NOT contaminated by caps re-charging pulses.
Putting those pulses onto the supply rails by doing the PSU tapping point incorrectly does not help with supply rail stability and makes correct implementation of local decoupling even more important.
I can't understand why Mauro broke convention and omitted full local decoupling.
I can't understand why the clones of MyRef continue to omit full local decoupling.
Leach tried to omit it and came back explaining why his Lo Tim oscillated. He explained why and gave us the solution: to add in 100uF of MF decoupling. HF decoupling alone was not good enough and made instability worse.
This is basic PCB layout stuff.
do you remember that we discussed the routing of the main smoothing caps link on the OTHER side of the PCB?
So far you are the only person to implement a PSU Zero Volts link to Main Audio Ground by using a single trace that is NOT contaminated by caps re-charging pulses.
Putting those pulses onto the supply rails by doing the PSU tapping point incorrectly does not help with supply rail stability and makes correct implementation of local decoupling even more important.
I can't understand why Mauro broke convention and omitted full local decoupling.
I can't understand why the clones of MyRef continue to omit full local decoupling.
Leach tried to omit it and came back explaining why his Lo Tim oscillated. He explained why and gave us the solution: to add in 100uF of MF decoupling. HF decoupling alone was not good enough and made instability worse.
This is basic PCB layout stuff.
Thanks - yes, I remember your suggestion and implemented the star ground almost exactly as you suggested, with the bridge-rectifier to cap current paths isolated from the audio-side ground current paths on different sides of the PCB, and connected with a single-point ground feed-through. This layout is also used in the MiniRef, IIRC.
Mauro probably found an alternative to stabilize the MyRef without the deleterious effects of the local rail-to-ground decoupling on sonics - that's the non-intuitive RC network at the input of the LM3886. I used the same to stabilize the MiniRef 3886.
The MiniRef also has the rail-to-ground small electrolytics C1 and C2 fairly close to the pins of at least one chipamp, and within a few cm of the other. It does use rail-to-rail decoupling almost right on top of the power supply pins of each chipamp.