I think that the two different layouts used by Andrew and Tom originate from two different causes of noise. Andrew is trying to reduce ground loop currents when more than one channel is used, this at a possible cost to audio quality.
Tom's approach is about designing the circuit to get best audio. The ground for a single amp has been designed to achieve low noise and low input errors. The measurements confirm low THD+N. It is going to be interesting to see how the new amp will preform in the real world at the hands of the DIY builder.
I annoy people at times. I don't want to change. I expect others to change.
Tom's approach is about designing the circuit to get best audio. The ground for a single amp has been designed to achieve low noise and low input errors. The measurements confirm low THD+N. It is going to be interesting to see how the new amp will preform in the real world at the hands of the DIY builder.
I annoy people at times. I don't want to change. I expect others to change.
The trouble is that if you make the ground loop high impedance, you increase the impact of capacitive coupling into that node. You also increase the error voltage. This hurts the overall performance of the amp.
The best design moves the ground loop out of the signal path. That's why my Modulus-86 has differential inputs. That way any error voltage developed on the ground reference will be attenuated by 90+ dB. It will be buried in the noise floor. In a single-ended input, the noise voltage will be in series with the input signal and see the full gain of the amplifier. Hence, for a single-ended system, the name of the game is to minimize the ground impedance (i.e. resistance and inductance).
Yep. That's exactly what I do.
I have pretty high confidence. I've built three of these circuits to date. They've all performed at the same high level. As long as the builder sticks to the BOM, I think they'll be fine. But you're right... It will be interesting to get some builder feedback.
I'm a leader. I drive change in others and strive for excellence...
At the same time, I also realize that this is DIY Audio and people volunteer their time here. Antek AS-2222 or AS-2224 transformer. KBPC2510 rectifier. 2x22000 uF, 35 or 50 V caps. I'm sure you, by doing a little homework yourself, can come up with a ±26~±28 V supply with those parts.
~Tom
IMHO a web page on simulation would be golden because it could very well drive much-needed adoption.
I like that about this thread as well. Thanks for the feedback. I have quite a bit more exploration to do on the LM3886. I won't promise to deliver it on a set schedule, but I would like to poke around some more to explore the limitations of the chip and how to work around them.
OK. I'll add it to the list...
~Tom
if using P2P or deadbug construction it's physically possible follow all your other suggestions but also to place the speaker return at the"decoupling ground blob". in that case is there reason nevertheless to segregate the speaker-return/feedback-ground/input-ref-ground from the blob or would one small "superblob" suffice.
If you can guarantee that no current flows between the speaker ground and the input/feedback ground, you can have a "super-blob". Note, however, that this "super-blob" will have something like ten pins or wires connecting to it, so I'm guessing you'll find it rather hard to manage in a P2P setup.
~Tom
~Tom
thanks.
understood.
in retrospect, i didn't ask quite the right question. what i have currently is 3-D P2P with a blob for the following: (1-5) PSU 0v, 4 chip bypass caps (.1 x7r 100v || 330uf Pan.FC 63v each with leads of about 0.5cm to the blob and the chip pins), (6) speaker return, and (7-8) two ~ 2 cm wires. one of the 2 wires goes to the mute gnd pin and the other goes to a "microblob" for the input ground and offset adjustment pot ground (it is a buffered inverting topo with a 1k input R and a 22k1 feedback R -- the buffer is an opa227 on a daughterboard, with its own isolated PSU (batteries currently), is capacitively coupled to its source and directly coupled to the chip input R).
so i'm mainly concerned that the 2 cm wire is a reasonable choice. i have see some setups where the wire is intentionally made longer (6-7cm) and i know AndrewT has expressed a preference for a low value R.
as an aside, i was an RF guy before i ventured into audio so short leads and microscopic ceramic caps and byzantine P2P and deadbug layouts don't bother me in the least.
Last edited:
the speaker return and hot are a twisted pair up to a spot between the blob and pin 3, where they separate for a length of about 1cm each to the blob and to pin 3, respectively. zoebel and thiele to come later to this same neighborhood.
I politely disagree. I'll explain below.
This I do agree with. That statement is contrary to your previous statement, however. Here's why: Transient current and HF current comes primarily from the local decoupling (22 uF low ESR + 1 uF X7R). Hence, the speaker return, Zobel network, and the local decoupling must all return to the SuperBlob.
The input and feedback ground can go to the speaker connector. You may want to shield this connection if it's routed a ways.
That may be, but to minimize HF THD, you need to keep that loop as low impedance as possible and with as little loop area as possible.
The speaker connections don't necessarily need to be twisted pair, though, that certainly doesn't hurt. The speaker out and return do need to be tightly coupled to minimize the loop area I mentioned above.
~Tom
i've always assumed that this is what AndrewT actually meant when he made a similar but overbroad assertion that if taken literally -- as you noted at the time -- would violate kirchoff's circuit laws.
Last edited:
this worked out as desired--dead quiet at wide open gain (23x).
thanks for the assistance.
The speaker cable and it's return do not carry VHF current changes. The local decoupling is a local ground for VHF current changes around the ClassAB output stage. The Zobel ONLY passes VHF and does need to connect to the local Decoupling Ground.
Some location/point near the output stage and near the input stage and near the input terminals and near the output terminals (that is 4 conditions) is made into the Main Audio Ground (MAG).
The MAG has wire connections to the PSU Zero Volts, to the local decoupling ground, to the Speaker Return, to the Signal Return ...
and can also connect to the Chassis either directly, or via a Disconnecting Network.
The Chassis connection can come to some other High Current cable elsewhere in the route including the PSU and MAG.
Note in the above para that the Speaker Return connects to the MAG directly. It is my contention that the speaker return does NOT need to connect DIRECTLY to the ultra low impedance of the local decoupling node. A short wire between MAG and local decoupling ground is acceptable. This is acceptable BECAUSE the speaker cable does not pass VHF, the Thiele Network prevents VHF signal passing around the speaker loop.
The difficult parts comes in two areas. Keeping those four conditions close in reality. That determines the layout of the amplifier relative to the IN/OUT sockets.
The second is maintaining the low Loop Area Pairs for all the Pairs that come into the MAG. That takes up space/volume. They cannot meet at a point. Some compromise must be made. The least compromise affects the VHF routes and pairs.
A bit more space/volume/length can be accommodated by the relatively slower speaker and PSU and Audio bandwidth signals.
Does your experimental data show that any of this is wrong?
Is there any need to extend the experimental data to show that a set of SHORT low LOOP AREA connections from MAG to IN socket, to OUT socket, to local Decoupling, rather than a ZERO length from all these to local Decoupling Ground, is mandatory?
Further, does it need to be proved that a longer, but low loop area, Power Triplet from PSU passing to the amplifier does not have significant deleterious effect on performance, or has your data proved this already?
Last edited: