Thank you.
The grounding layout of the Modulus-86 and Parallel-86 resulted from extensive simulation of the grounding network (impedance of traces and planes). The simulations were backed up by experiments on the first PCB prototype. As Twest points out, in precision circuit design, everything matters.
A good place to start for a bit of insight is the LM3886 point-to-point vs PCB thread. Specifically, Post #30 shows quite a bit of data.
Since you have both Modulus-86, Parallel-86, and the capability to measure them, you could design an experiment to test out your hypothesis. I suggest measuring THD+N vs Frequency at the max output power as function of load ground return path. If you can improve on the performance of the Modulus/Parallel-86, I'd be rather surprised - and very interested in the data.
The biggest difference will be at higher frequencies (>5 kHz). Depending on your test equipment's capabilities, you may be able to quantify the difference in the grounding layouts from a spot measurement of the THD at 10 kHz or 20 kHz. In the THD vs freq. sweep, the difference is usually rather obvious (as seen in Post #30 linked to above). Beware that measuring vanishingly low THD+N requires as much care as designing the circuit to deliver that performance level. Just make sure you actually measure the circuit and not your test equipment (been there, done that).
Tom
We agree so far, but I thought I'd expand a bit on your statement. There are two sides to this:
1) Feedback tap point
2) Feedback ground reference
You cover 1) nicely. To address 2), the feedback ground reference should be at the speaker connector GND connection. Further, the ground network for the control circuitry must be designed such that only minimal current flows in this reference ground connection (minimize the error voltage).
One could "geek out" and perform differential sensing of the output voltage. I'll leave that as an exercise for the reader.
Tom
Tom,
I have been having a ground loop issue using the MiniDSP with a chipamp build for the LXMini with sub. In total I have 6 channels and two MiniDSP units and don't know what to do to eliminate ground loops and hum. I am no where close to your experience and would like to know any details of what you did to eliminate the ground loops using the MiniDSP units. Could it be the CHIPAMP Design? I built three identical amps in separate chassis for the application. The amps work very well when connected to stereo speakers without an ground loop issues, but when connected to the MiniDSP I always have a loop and hum. Any advice would be appreciated.
Thanks,
Tim
I have been having a ground loop issue using the MiniDSP with a chipamp build for the LXMini with sub. In total I have 6 channels and two MiniDSP units and don't know what to do to eliminate ground loops and hum. I am no where close to your experience and would like to know any details of what you did to eliminate the ground loops using the MiniDSP units. Could it be the CHIPAMP Design? I built three identical amps in separate chassis for the application. The amps work very well when connected to stereo speakers without an ground loop issues, but when connected to the MiniDSP I always have a loop and hum. Any advice would be appreciated.
Thanks,
Tim
That seems like a topic worth its own thread. Should you choose to start one, I'll be happy to help to the extent my time allows.
You can't avoid ground loops, but you can reduce their impact. One way to accomplish this is to use differential signaling. That's why all my amps have differential inputs. Another way to lower the impact of the ground loops is to minimize the loop area.
In my amp, I didn't do anything special to combat the ground loops caused by the MiniDSP. I did my best to minimize the loop area and I used pseudo-differential signaling between the MiniDSP and the Modulus-86. Also note that I'm connecting to a Sonos Connect which doesn't have a safety ground (two-prong mains connection). This helps to minimize the ground issues as it eliminates the ground loop through the mains connections.
Tom
Try the following:
- Disconnect the earth connections from the power input on the chip amp pcb's
- Disconnect the earth connections from the speaker output on the chip amp pcb's and connect these to the star ground close to the power supply
- Connect the earth connection of the MiniDsp to the same star ground.
- Separate the decoupling caps on the chip amp from ground so that they connect directly between the two power rails.
The chip amp needs a ground connection for references, but draws little power. This current can easily be provided by the ground connection of the signal input, which will be connected through the minidsp to the star ground.
By doing so, your ground will follow the signal and a ground loop becomes impossible.
As I mentioned before in this thread, using the chip amp pcb for the load return current is a bad idea.
Last edited:
That will increase the supply impedance dramatically and lead to extremely poor performance. You're basically forcing all supply return current to flow through the input ground at that point. That's a rather reckless recommendation.
That you can do. It won't do anything to address the ground loop involving the MiniDSP and will lead to a rise in THD starting in the low kHz range. This is likely to result in the common "chip amp strained sound".
Good luck with that considering that the MiniDSP routes all grounds to the chassis. I suppose you could float the MiniDSP. One could use the plastic press fit standoffs for mounting the MiniDSP board to the chassis. That might work.
This breaks the local decoupling on the amp PCB, which is likely to cause instability on the LM3886. That's another reckless recommendation.
Except for the load return current. Where do you suppose that comes from?
Except for the load return current, that's true. However, connecting the chip amp ground to a star ground will dork up the THD. I posted measurements in another thread about a year ago. I provided you with a link to it a few pages back. In case you missed, here it is again: LM3886 P2P vs PCB Post #30.
Really? How about the ground loop formed between the source and the amp?
Would you mind backing that up with data? You previously indicated that you have test equipment available to you so it should be possible for you to do so.
Tom
As an addendum, I am working with standard LM3886's on well executed boards, and I have the impression Claudman is also using something like that. I am in the middle of a big project, so will need to come back in more detail on this issue at a later stage. It may be counterintuitive at first, but it works.
Look... I've presented measured data that comes from a well-executed LM3886 PCB in the "LM3886 P2P vs PCB" thread. I've linked to it at least twice within the past couple of pages in this thread. It's unfortunate for you that measured data doesn't match your preconceived opinion. You're obviously free to connect your LM3886 however you want, but I would prefer that any recommendations on the connection of a Modulus-86 or Parallel-86 are rooted in facts rather than speculation.
On the topic of decoupling: The instantaneous load current is provided by the decoupling and local reservoir caps on the PCB. This current flows from the decoupling/reservoir cap through the VCC pin to the OUT pin through the load back to the decoupling/local reservoir cap. If you take this current flow and route it to a common star point off the PCB, the ground route gets considerably longer. This increases the inductance of the supply network. This, in turn, ruins the THD at high frequency. I have measured this and provided you with the data. If high THD is what you want, great! Go ahead and connect your LM3886 that way. However, if you want mediocre performance, why build a Modulus-86? The whole point of the Modulus-86 is to deliver high performance. Best in class performance. The Modulus-86 delivers this when connected as shown in the design documentation.
As for the ground loops caused by the MiniDSP: Of course you can float the MiniDSP. Just use plastic hardware to mount it and keep the input connectors isolated from the chassis. Of course, its ground reference will need to meet the signal ground of the Modulus-86 at one point. I never suggested that you rely on a single wire for the signal transfer. Current flows in loops. This requires both a forward conductor and a return conductor.
Tom
On the topic of decoupling: The instantaneous load current is provided by the decoupling and local reservoir caps on the PCB. This current flows from the decoupling/reservoir cap through the VCC pin to the OUT pin through the load back to the decoupling/local reservoir cap. If you take this current flow and route it to a common star point off the PCB, the ground route gets considerably longer. This increases the inductance of the supply network. This, in turn, ruins the THD at high frequency. I have measured this and provided you with the data. If high THD is what you want, great! Go ahead and connect your LM3886 that way. However, if you want mediocre performance, why build a Modulus-86? The whole point of the Modulus-86 is to deliver high performance. Best in class performance. The Modulus-86 delivers this when connected as shown in the design documentation.
As for the ground loops caused by the MiniDSP: Of course you can float the MiniDSP. Just use plastic hardware to mount it and keep the input connectors isolated from the chassis. Of course, its ground reference will need to meet the signal ground of the Modulus-86 at one point. I never suggested that you rely on a single wire for the signal transfer. Current flows in loops. This requires both a forward conductor and a return conductor.
Tom
I'm looking forward to seeing the data. Measure THD+N vs frequency at max output power. You'll notice a dramatic rise in THD+N above a few kHz. You'll be at least 20 dB if not 40 dB higher THD+N at 20 kHz than the LM3886 data sheet spec with the wiring scheme you suggest.
Feel free to prove me wrong, but please follow my example and back your claims up with data. You have the gear.... I do understand that time comes at a premium. It does for me too.
Tom
Build first, measure later, eh? My experience is similar to Tom's in suggesting 30 to 40dB worse than the Mod and Parallel is likely. This would indeed fall in the workable range but describing it as "good results" and advocating adoption as an improvement prior to verification is, at best, premature.
Most assertions in post 932 are suboptimal or incorrect due to RLCG and other error terms being mistakenly treated as negligible. In particular you appear to be struggling with correctly accounting for inductance; that may be a useful area to prioritize for improvement.
No, of course I measured, but not in a publishable form.
As to what Tom posted in #934: the decoupling and local reservoir caps are, as the word implies, capacitors. The only thing that matter is that they can push against a hard rail, the voltage of which is unimportant. It can be ground, it can be another power rail. Since class AB only draws instantaneous current from one line at the time, you can use the power lines as terminals for the decoupling caps.
Just two more things.
The issue at hand is how to connect different LM3886's to different outputs from a device with common grounding. What Tom describes are tests with only 1 device. The problem with the plurality of LM3886's connected to a common ground with the load current flowing through the LM3886 PCB's is that different ground potentials develop amongst the different 3886 PCB's. As I mentioned before, with a balanced input, this matters less (because these fluctuating ground potentials are seen as common mode signals) than with unbalanced connection, but it still does somewhat.
Now, as to "If you take this current flow and route it to a common star point off the PCB, the ground route gets considerably longer." The fallacy in this is that the load will typically be connected by longish loudspeaker cords anyways. The couple of additional inches inside the amplifier are really of no consequence.
As to what Tom posted in #934: the decoupling and local reservoir caps are, as the word implies, capacitors. The only thing that matter is that they can push against a hard rail, the voltage of which is unimportant. It can be ground, it can be another power rail. Since class AB only draws instantaneous current from one line at the time, you can use the power lines as terminals for the decoupling caps.
Just two more things.
The issue at hand is how to connect different LM3886's to different outputs from a device with common grounding. What Tom describes are tests with only 1 device. The problem with the plurality of LM3886's connected to a common ground with the load current flowing through the LM3886 PCB's is that different ground potentials develop amongst the different 3886 PCB's. As I mentioned before, with a balanced input, this matters less (because these fluctuating ground potentials are seen as common mode signals) than with unbalanced connection, but it still does somewhat.
Now, as to "If you take this current flow and route it to a common star point off the PCB, the ground route gets considerably longer." The fallacy in this is that the load will typically be connected by longish loudspeaker cords anyways. The couple of additional inches inside the amplifier are really of no consequence.
It would therefore follow the return path within amplifier chassis is also of no consequence. Resolving this contradiction within your assertions seems desirable.
It would therefore follow the return path within amplifier chassis is also of no consequence. Resolving this contradiction within your assertions seems desirable.
No, it is not the return path through the chassis, but the return part through the PCB what I am talking about. But enough said, when I get the time to do so I will make some measurements of my own and let's take it from there.
Best,
va
The output from the chipamp sees two loads, the Zobel and the external load (the speaker) in series with the output inductor.
The Zobel only presents a high frequency load. Typically >200kHz
The speaker presents a low and some mid frequency loading. Typically around 100kHz. Anything above that tries to find an easier route, so 100 to 200KHz will share the Zobel and the speaker route.
The very fastest changes in current will go through the Zobel. It is the Zobel that must be returned to the decoupling ground.
Last edited: