Originally Posted by mikeaj
With a particular switchbox (if the ground of the "unused" source is not floated) you could still readily verify if sloppy EMC design will make a measurable or audible difference in practice, by testing it with and without. Have you?
Certainly there should be no significant high-frequency content (above half the sampling rate) in the signal from the source, unless it has severe issues, which is not impossible. I'm by no means an expert in EMC, but we're by no means talking about the usual high-frequency systems where such considerations are more important. Where do you suggest there is significant coupling, between what?
Don't know how to split quote on this forum...this is to the first Q, the second should extrude from the first answer.
Q1...Yes, I have. The most drastic one I had had this:
The input RCA was on the back panel farthest left from front view. the line cord was on the far right. The power supply star ground was approximately in the middle of the chassis, on the bottom. The speaker outputs were next to the line cord entry point. All the pass transistors for the output were linear across the back of the chassis, taking up the entire length, it was a 3 ru panel.
The input pair was on the far left, so the input shielded cable was only about 5 inches long.
In practice, the following problems reared their ugly head.
1. Any ground loop currents flow along the back panel of the chassis, from the rca shield to the line cord ground. For 60 hz currents, the aluminum basically can be considered as a wide flat conductor, as eddy currents have only started to constrict the current path that a DC current would take. As the frequency of the loop current increases, the path of the current will become more constricted as the current attempts to take the path of least reactance.
2. The feedback network which sets both the DC fback gain as well as the hf gain, had a star ground back to the cap bank. This is a different reference point than the connection of the input jack at the rca. The two paths form a loop of ground conductor, and this loop was significant enough to trap flux caused by currents along the back panel.
3. The output devices as wired, did not control the current path of the supply rails nor the current path of the output run. Done properly, the currents should have all been twisted together to drop the mangetic field caused by the outputs. They were NOT, and as a consequence, allowed the ground loop formed internal to the chassis to couple to the output currents.
As a direct consequence of these design flaws, the amplifier allowed both ground loop currents and internal power draw currents to couple into the analog path, by compromising the star grounding reference points, and more heinous, some of the fields managed to couple directly into the input pair on the pc board. This is an avenue of ingress which has the FULL GAIN of the amplifier circuitry.
As a result, the amplifier was susceptible to line cord selection both because of cord geometry as well as grounding resistance variations. It was also susceptible to coupling between the IC shield and the total ground loop formed by the source, amp, and the wires. That susceptibility included a phase margin which changed depending on the cords. One time, it was sufficiently low that a spike in a record caused both amplifiers in the system to break into oscillation somewhere below the unity gain frequency, and it exploded the crossover caps in both speaker columns.
Q2. The higher the frequency, the less of a problem you have with direct current path issues. The reason is, for a coax input at very high frequencies, the ground loop inductive reactance far exceeds the impedance the signal would take through the coax..so the signal will return via the shield. When that occurs, the shield is effective.
ps..Forgot to add. On the test bench where I checked the amplifier for distortion, there was no ground loop formed by the input (blame good test gear), so the amp read perfectly well into 4 ohms at 300 watts resistive. And, being resistive, the output draw was only two quadrant..a reactive load swaps the magfield drive loop in rather strange and wonderful ways..
Edited by jnjn - 12/28/11 at 3:40pm